IMW 2020: Leaderboard

Current version: aaba1df9 (2020-09-17, 22:53 UTC)

Categories are non-exclusive: e.g. small descriptors compete with larger descriptors, but not vice-versa. For the sake of clarity, we do separate 2k and 8k in the leaderboards. Please note that this is a static website: you may want to force a reload if it does not update properly. We are working on new features and tracks: the organizers reserve the right to update terms and conditions within the next few weeks.

Prize #1: unlimited keypoints (8k) / standard descriptors (512 bytes)


Prize #2: restricted keypoints (2k) / standard descriptors (512 bytes)



Phototourism: unlimited keypoints / standard descriptors (512 bytes)

Note: entries with the same multi-view configuration may seem duplicated. This is normal: performance is averaged across tasks.
Stereo Multiview Avg.
Method NF NI Rep.
(3 pix.)
MS
(3 pix.)
mAA
(at 10o)
NM NL TL ATE mAA
(at 100)
mAA
(at 100)
Authors Keypoint Descriptor Summary Paper Website Processing date Flags
Submission ID: 00078
SOSNet-Upright w/ MAGSAC (no FLA...
Size: 512 bytes. Matches: built-in
7829.63 679.94
Rank: 17/127
0.486
Rank: 44/127
0.855
Rank: 30/127
0.56626
(±0.00020)
Rank: 23/127
600.23
Rank: 34/127
4765.43
Rank: 26/127
4.584
Rank: 27/127
0.403
Rank: 26/127
0.73027
(±0.00284)
Rank: 25/127
0.64826
Rank: 26/127
Challenge organizers (contact) sift8k sosnet-upright (128 float32: 512 bytes) Upright SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00597
SIFT8k-HardNet64
Size: 512 bytes. Matches: built-in
6589.88 381.75
Rank: 83/127
0.467
Rank: 107/127
0.852
Rank: 36/127
0.52511
(±0.00056)
Rank: 52/127
420.42
Rank: 76/127
3595.93
Rank: 80/127
4.496
Rank: 47/127
0.417
Rank: 42/127
0.70738
(±0.00218)
Rank: 48/127
0.61625
Rank: 45/127
caoliang (contact) sift8k hardnet (128 float32: 512 bytes) SIFT up to 8000 keypoints, hardnet extract descriptors.Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-05-27 is_submission, is_challenge_2020
Submission ID: 00042
ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
7150.21 161.98
Rank: 124/127
0.514
Rank: 16/127
0.653
Rank: 115/127
0.16159
(±0.00090)
Rank: 118/127
910.31
Rank: 13/127
1423.38
Rank: 126/127
2.722
Rank: 126/127
0.897
Rank: 127/127
0.08054
(±0.00242)
Rank: 127/127
0.12106
Rank: 127/127
Challenge organizers (contact) orb orb (32 uint8: 32 bytes) ORB with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00112
VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 204.37
Rank: 120/127
0.547
Rank: 12/127
0.823
Rank: 87/127
0.36954
(±0.00081)
Rank: 110/127
347.39
Rank: 109/127
3209.10
Rank: 101/127
4.126
Rank: 105/127
0.517
Rank: 114/127
0.58657
(±0.00120)
Rank: 113/127
0.47806
Rank: 112/127
Challenge organizers (contact) hessian vlsift (128 float32: 512 bytes) VL-Hess-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00508
modified-r2d2
Size: 512 bytes. Matches: built-in
7791.82 160.64
Rank: 125/127
0.517
Rank: 15/127
0.785
Rank: 103/127
0.32815
(±0.00037)
Rank: 111/127
328.39
Rank: 115/127
2811.27
Rank: 113/127
4.250
Rank: 91/127
0.488
Rank: 106/127
0.62899
(±0.00193)
Rank: 102/127
0.47857
Rank: 111/127
caoliang (contact) modified-r2d2 modified-r2d2 (128 float32: 512 bytes) r2d2 with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value to increase performance. https://europe.naverlabs.com/wp-content/uploads/2019/09/R2D2-Repeatable-and-Reliable-Detector-and-Descriptor-2.pdf https://github.com/naver/r2d2 20-04-25 is_submission, is_challenge_2020
Submission ID: 00536
HarrisZ + DeepOrientation + SOSN...
Size: 512 bytes. Matches: custom
2410.18 1164.84
Rank: 3/127
0.368
Rank: 118/127
0.341
Rank: 125/127
0.00607
(±0.00000)
Rank: 126/127
1191.31
Rank: 4/127
2215.65
Rank: 118/127
4.014
Rank: 116/127
0.524
Rank: 116/127
0.58741
(±0.00250)
Rank: 112/127
0.29674
Rank: 120/127
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] N/A N/A 20-05-01 is_submission, is_challenge_2020
Submission ID: 00019
Upright Root-SIFT (OpenCV), DEGE...
Size: 512 bytes. Matches: built-in
7829.24 355.69
Rank: 93/127
0.487
Rank: 23/127
0.844
Rank: 55/127
0.51009
(±0.00033)
Rank: 69/127
548.18
Rank: 41/127
4406.41
Rank: 45/127
4.385
Rank: 59/127
0.442
Rank: 70/127
0.68859
(±0.00254)
Rank: 72/127
0.59934
Rank: 68/127
Challenge organizers (contact) sift-lowth rootsift-upright (128 float32: 512 bytes) Upright Root-SIFT with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00085
LogPolar-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 611.19
Rank: 26/127
0.486
Rank: 44/127
0.830
Rank: 75/127
0.51287
(±0.00038)
Rank: 66/127
483.93
Rank: 61/127
4405.57
Rank: 46/127
4.542
Rank: 39/127
0.423
Rank: 52/127
0.70680
(±0.00247)
Rank: 49/127
0.60983
Rank: 60/127
Challenge organizers (contact) sift8k logpolar96-fixed-upright (128 float32: 512 bytes) Upright LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-23 is_baseline
Submission ID: 00571
ContextDesc Upright + OANetV2 + ...
Size: 512 bytes. Matches: custom
7830.09 827.45
Rank: 9/127
0.487
Rank: 25/127
0.775
Rank: 104/127
0.53810
(±0.00000)
Rank: 43/127
855.38
Rank: 16/127
6561.52
Rank: 3/127
4.522
Rank: 41/127
0.370
Rank: 8/127
0.76702
(±0.00081)
Rank: 6/127
0.65256
Rank: 23/127
Jiahui Zhang, Zixin Luo, Hongkai Chen (contact) contextdesc-upright contextdesc-upright (128 float32: 512 bytes) ContextDesc with 8000 SIFT features, using improved OANet matcher and DEGENSAC post-processing N/A N/A 20-05-21 is_submission, is_challenge_2020
Submission ID: 00709
DISK (LCC/depth)
Size: 512 bytes. Matches: built-in
7844.17 1238.52
Rank: 2/127
0.644
Rank: 1/127
0.852
Rank: 37/127
0.55847
(±0.00084)
Rank: 27/127
1663.81
Rank: 1/127
7483.98
Rank: 2/127
5.922
Rank: 1/127
0.391
Rank: 23/127
0.75024
(±0.00316)
Rank: 15/127
0.65435
Rank: 21/127
Under review! NeurIPS anonymous submission, ID 1194 (To be released: 21-06-02) disk-cc-continued-20-imsize-1024-nms-3-nump-8000 disk-cc-continued-20-imsize-1024-nms-3-nump-8000 (128 float32: 512 bytes) Local feature model learned via policy gradient. Model trained with a cycle-consistency loss and supervised with depth. Trained on MegaDepth, removing conflicts with the test data. For inference, images are resized to 1024 pixels on the longest edge, with NMS over a 3x3 window. We take the top 8000 features by score. N/A N/A 20-06-02 is_submission, is_challenge_2020, is_under_review
Submission ID: 00573
Guided-HardNet-qht
Size: 512 bytes. Matches: custom
7829.63 887.83
Rank: 7/127
0.486
Rank: 44/127
0.774
Rank: 105/127
0.55224
(±0.00000)
Rank: 31/127
538.28
Rank: 43/127
4693.85
Rank: 29/127
4.650
Rank: 9/127
0.381
Rank: 11/127
0.75930
(±0.00177)
Rank: 8/127
0.65577
Rank: 20/127
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet-qht (128 float32: 512 bytes) sift and hardnet with 8k features, using the modified guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-23 is_submission, is_challenge_2020
Submission ID: 00010
AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
7857.11 246.74
Rank: 115/127
0.553
Rank: 9/127
0.735
Rank: 111/127
0.30717
(±0.00122)
Rank: 114/127
479.55
Rank: 65/127
2778.68
Rank: 114/127
3.393
Rank: 124/127
0.737
Rank: 126/127
0.36048
(±0.00382)
Rank: 125/127
0.33383
Rank: 116/127
Challenge organizers (contact) akaze-lowth akaze (61 uint8: 61 bytes) AKAZE with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00068
GeoDesc-Upright w/ DEGENSAC (no ...
Size: 512 bytes. Matches: built-in
7829.63 409.87
Rank: 75/127
0.486
Rank: 44/127
0.869
Rank: 17/127
0.52674
(±0.00049)
Rank: 50/127
458.60
Rank: 66/127
4146.83
Rank: 57/127
4.412
Rank: 55/127
0.431
Rank: 61/127
0.70442
(±0.00189)
Rank: 53/127
0.61558
Rank: 46/127
Challenge organizers (contact) sift8k geodesc-upright (128 float32: 512 bytes) Upright GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00532
affnet-hardnet
Size: 512 bytes. Matches: built-in
7925.53 226.75
Rank: 117/127
0.452
Rank: 113/127
0.649
Rank: 117/127
0.31958
(±0.00055)
Rank: 112/127
243.28
Rank: 124/127
3155.72
Rank: 104/127
3.595
Rank: 121/127
0.554
Rank: 117/127
0.52578
(±0.00343)
Rank: 117/127
0.42268
Rank: 114/127
caoliang (contact) affnet hardnet (128 float32: 512 bytes) affnet up to 8000 keypoints, harnet extract descriptors.Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo https://arxiv.org/abs/1711.06704 https://github.com/ducha-aiki/affnet 20-04-24 is_submission, is_challenge_2020
Submission ID: 00071
L2-Net w/ DEGENSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 366.00
Rank: 88/127
0.472
Rank: 86/127
0.852
Rank: 40/127
0.52953
(±0.00043)
Rank: 49/127
339.00
Rank: 113/127
3424.94
Rank: 90/127
4.206
Rank: 97/127
0.452
Rank: 85/127
0.66445
(±0.00213)
Rank: 87/127
0.59699
Rank: 71/127
Challenge organizers (contact) sift8k l2net (128 float32: 512 bytes) L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-23 is_baseline
Submission ID: 00024
SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
7860.73 238.66
Rank: 116/127
0.472
Rank: 81/127
0.824
Rank: 85/127
0.45426
(±0.00097)
Rank: 99/127
418.86
Rank: 78/127
3515.63
Rank: 87/127
4.001
Rank: 117/127
0.502
Rank: 109/127
0.60193
(±0.00185)
Rank: 110/127
0.52810
Rank: 102/127
Challenge organizers (contact) sift-lowth sift (128 float32: 512 bytes) SIFT with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00084
LogPolar-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 459.75
Rank: 61/127
0.486
Rank: 44/127
0.852
Rank: 38/127
0.52316
(±0.00034)
Rank: 57/127
483.93
Rank: 61/127
4405.57
Rank: 46/127
4.542
Rank: 39/127
0.422
Rank: 49/127
0.70680
(±0.00247)
Rank: 49/127
0.61498
Rank: 48/127
Challenge organizers (contact) sift8k logpolar96-fixed-upright (128 float32: 512 bytes) Upright LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00506
sift and hardnet64
Size: 512 bytes. Matches: built-in
7861.62 358.55
Rank: 91/127
0.472
Rank: 84/127
0.849
Rank: 43/127
0.52085
(±0.00009)
Rank: 58/127
585.33
Rank: 36/127
4957.86
Rank: 24/127
4.294
Rank: 83/127
0.419
Rank: 46/127
0.70472
(±0.00257)
Rank: 52/127
0.61279
Rank: 53/127
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64 (128 float32: 512 bytes) SIFT with 8000 keypoints, hardnet64 with 128 descriptors N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00100
KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 448.11
Rank: 64/127
0.582
Rank: 2/127
0.838
Rank: 62/127
0.39971
(±0.00147)
Rank: 105/127
356.21
Rank: 100/127
3366.01
Rank: 93/127
4.319
Rank: 77/127
0.462
Rank: 94/127
0.64829
(±0.00174)
Rank: 95/127
0.52400
Rank: 103/127
Challenge organizers (contact) keynettuned vlhardnet (128 float32: 512 bytes) KeyNet-HardNet with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00569
Upright-SIFT-HardNet
Size: 512 bytes. Matches: built-in
6667.86 376.15
Rank: 85/127
0.491
Rank: 18/127
0.844
Rank: 53/127
0.51560
(±0.00043)
Rank: 63/127
397.09
Rank: 87/127
3417.56
Rank: 92/127
4.479
Rank: 50/127
0.448
Rank: 84/127
0.68803
(±0.00532)
Rank: 73/127
0.60181
Rank: 66/127
caoliang (contact) sift8k hardnet (128 float32: 512 bytes) SIFT up to 8000 keypoints, harnet extract descriptors.Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-05-20 is_submission, is_challenge_2020
Submission ID: 00066
SOSNet-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 537.85
Rank: 43/127
0.486
Rank: 44/127
0.841
Rank: 58/127
0.53531
(±0.00061)
Rank: 45/127
521.37
Rank: 51/127
4473.48
Rank: 40/127
4.597
Rank: 23/127
0.414
Rank: 38/127
0.72179
(±0.00330)
Rank: 34/127
0.62855
Rank: 40/127
Challenge organizers (contact) sift8k sosnet-upright (128 float32: 512 bytes) Upright SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00076
SOSNet w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 563.27
Rank: 41/127
0.472
Rank: 86/127
0.846
Rank: 47/127
0.55170
(±0.00042)
Rank: 32/127
508.66
Rank: 56/127
4502.19
Rank: 36/127
4.377
Rank: 68/127
0.405
Rank: 32/127
0.71820
(±0.00298)
Rank: 39/127
0.63495
Rank: 34/127
Challenge organizers (contact) sift8k sosnet (128 float32: 512 bytes) SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00065
SOSNet-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 407.29
Rank: 76/127
0.486
Rank: 44/127
0.864
Rank: 26/127
0.54098
(±0.00086)
Rank: 39/127
521.37
Rank: 51/127
4473.48
Rank: 40/127
4.597
Rank: 23/127
0.417
Rank: 40/127
0.72179
(±0.00330)
Rank: 34/127
0.63139
Rank: 36/127
Challenge organizers (contact) sift8k sosnet-upright (128 float32: 512 bytes) Upright SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00522
SIFT and HardNet64 train scale(5...
Size: 512 bytes. Matches: built-in
7830.09 513.57
Rank: 50/127
0.486
Rank: 32/127
0.852
Rank: 41/127
0.55385
(±0.00076)
Rank: 30/127
822.07
Rank: 18/127
5660.19
Rank: 15/127
4.555
Rank: 34/127
0.404
Rank: 30/127
0.72744
(±0.00455)
Rank: 28/127
0.64065
Rank: 31/127
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-raw64 (128 float32: 512 bytes) SIFT with 8000 keypoints(raw 64), hardnet64 with 128 descriptors, FLANN disabled N/A N/A 20-04-26 is_submission, is_challenge_2020
Submission ID: 00554
ContextDesc-Upright w/ DEGENSAC ...
Size: 512 bytes. Matches: built-in
7830.09 470.69
Rank: 59/127
0.487
Rank: 25/127
0.877
Rank: 3/127
0.55697
(±0.00079)
Rank: 28/127
543.45
Rank: 42/127
4543.76
Rank: 35/127
4.498
Rank: 46/127
0.405
Rank: 31/127
0.72902
(±0.00255)
Rank: 27/127
0.64299
Rank: 30/127
Zixin Luo, Jiahui Zhang, Hongkai Chen (contact) sift-def contextdesc-upright (128 float32: 512 bytes) ContextDesc with 8000 SIFT features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value and use only single orientation to increase performance. FLANN disabled. https://arxiv.org/abs/1904.04084 https://github.com/lzx551402/contextdesc 20-05-11 is_submission, is_challenge_2020
Submission ID: 00504
ASLFeat (MS) w/ DEGENSAC (no FLA...
Size: 512 bytes. Matches: built-in
6948.83 390.79
Rank: 79/127
0.550
Rank: 10/127
0.797
Rank: 100/127
0.46102
(±0.00067)
Rank: 96/127
510.80
Rank: 53/127
3817.14
Rank: 73/127
4.577
Rank: 29/127
0.444
Rank: 75/127
0.68246
(±0.00173)
Rank: 76/127
0.57174
Rank: 88/127
Zixin Luo, Jiahui Zhang (contact) aslfeat aslfeat (128 float32: 512 bytes) ASLFeat (MS) with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC. FLANN disabled. https://arxiv.org/abs/2003.10071 https://github.com/lzx551402/ASLFeat 20-04-25 is_submission, is_challenge_2020
Submission ID: 00548
HarrisZ + DeepOrientation + SOSN...
Size: 512 bytes. Matches: custom
2410.18 343.78
Rank: 97/127
0.368
Rank: 118/127
0.607
Rank: 119/127
0.28795
(±0.00000)
Rank: 115/127
351.53
Rank: 103/127
1699.08
Rank: 123/127
4.124
Rank: 108/127
0.507
Rank: 110/127
0.61370
(±0.00246)
Rank: 106/127
0.45082
Rank: 113/127
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 0.75, degeneracy check true) [Mishkin 2019] N/A N/A 20-05-03 is_submission, is_challenge_2020
Submission ID: 00519
SOSNet-Upright-AdaLAM
Size: 512 bytes. Matches: custom
7830.09 803.50
Rank: 11/127
0.486
Rank: 32/127
0.818
Rank: 92/127
0.56056
(±0.00000)
Rank: 25/127
827.85
Rank: 17/127
6157.62
Rank: 10/127
4.700
Rank: 5/127
0.381
Rank: 12/127
0.75917
(±0.00446)
Rank: 9/127
0.65987
Rank: 18/127
Luca Cavalli, Viktor Larsson, Martin Oswald, Torsten Sattler, Marc Pollefeys (contact) sift-def sosnet-upright (128 float32: 512 bytes) Using upright SOSNet descriptors with 8000 features, nearest neighbor matching and outlier rejection enforcing local affine consistency within a confidence-based adaptive error tolerance. Matches post-processed with DEGENSAC. https://arxiv.org/abs/2006.04250 N/A 20-04-26 is_submission, is_challenge_2020
Submission ID: 00101
KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 598.28
Rank: 32/127
0.582
Rank: 2/127
0.826
Rank: 82/127
0.49856
(±0.00038)
Rank: 75/127
356.21
Rank: 100/127
3366.01
Rank: 93/127
4.319
Rank: 77/127
0.464
Rank: 97/127
0.64829
(±0.00174)
Rank: 95/127
0.57342
Rank: 86/127
Challenge organizers (contact) keynettuned vlhardnet (128 float32: 512 bytes) KeyNet-HardNet with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00501
Guided matching with Upright roo...
Size: 512 bytes. Matches: custom
7829.27 609.35
Rank: 27/127
0.486
Rank: 29/127
0.786
Rank: 102/127
0.49072
(±0.00000)
Rank: 79/127
1363.98
Rank: 3/127
7835.75
Rank: 1/127
4.219
Rank: 96/127
0.454
Rank: 87/127
0.68064
(±0.00390)
Rank: 79/127
0.58568
Rank: 77/127
(contact) sift upright-root-sift (128 float32: 512 bytes) In submission N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00060
L2-Net w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 386.81
Rank: 80/127
0.472
Rank: 86/127
0.813
Rank: 96/127
0.49054
(±0.00043)
Rank: 80/127
314.54
Rank: 119/127
3314.19
Rank: 99/127
4.204
Rank: 99/127
0.465
Rank: 98/127
0.65665
(±0.00287)
Rank: 89/127
0.57360
Rank: 85/127
Challenge organizers (contact) sift8k l2net (128 float32: 512 bytes) L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-22 is_baseline
Submission ID: 00545
structured feature description
Size: 512 bytes. Matches: built-in
7830.21 411.68
Rank: 73/127
0.486
Rank: 42/127
0.865
Rank: 23/127
0.54442
(±0.00084)
Rank: 37/127
524.05
Rank: 49/127
4467.23
Rank: 42/127
4.551
Rank: 35/127
0.412
Rank: 35/127
0.72729
(±0.00300)
Rank: 29/127
0.63586
Rank: 33/127
Mahdi Abolfazli Esfahani, Han Wang (contact) sift sfd (128 float32: 512 bytes) structured descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC N/A N/A 20-04-29 is_submission, is_challenge_2020
Submission ID: 00120
CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 265.75
Rank: 112/127
0.472
Rank: 78/127
0.873
Rank: 11/127
0.46073
(±0.00062)
Rank: 97/127
356.56
Rank: 97/127
3550.63
Rank: 82/127
4.275
Rank: 85/127
0.439
Rank: 67/127
0.68875
(±0.00138)
Rank: 69/127
0.57474
Rank: 84/127
Challenge organizers (contact) sift8k hardnetamos (128 float32: 512 bytes) CV-DoG-HardNetAmos with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00568
Guided-SOSNet-lib-p
Size: 512 bytes. Matches: custom
7829.63 508.45
Rank: 51/127
0.486
Rank: 44/127
0.874
Rank: 10/127
0.57982
(±0.00000)
Rank: 16/127
524.66
Rank: 48/127
4618.86
Rank: 32/127
4.632
Rank: 16/127
0.369
Rank: 6/127
0.75888
(±0.00437)
Rank: 10/127
0.66935
Rank: 9/127
Chen Shen, Zhipeng Wang, Jun Zhang, Zhiwei Ruan, Zhongkun Chen, Jingchao Zhou, Pengfei Xu (contact) sift8k sosnet-lib-p (128 float32: 512 bytes) sift and sosnet with 8k features, using the guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-15 is_submission, is_challenge_2020
Submission ID: 00051
HardNet w/ DEGENSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 432.32
Rank: 70/127
0.472
Rank: 86/127
0.866
Rank: 22/127
0.55430
(±0.00031)
Rank: 29/127
426.84
Rank: 74/127
4001.40
Rank: 65/127
4.339
Rank: 75/127
0.419
Rank: 44/127
0.70962
(±0.00191)
Rank: 46/127
0.63196
Rank: 35/127
Challenge organizers (contact) sift8k hardnet (128 float32: 512 bytes) HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00518
UprightRootSIFT-AdaLAM
Size: 512 bytes. Matches: custom
6449.42 435.53
Rank: 69/127
0.436
Rank: 115/127
0.825
Rank: 84/127
0.49625
(±0.00000)
Rank: 76/127
449.49
Rank: 69/127
4353.56
Rank: 48/127
4.453
Rank: 53/127
0.397
Rank: 24/127
0.72490
(±0.00190)
Rank: 30/127
0.61057
Rank: 57/127
Luca Cavalli, Viktor Larsson, Martin Oswald, Torsten Sattler, Marc Pollefeys (contact) sift-def rootsift-upright (128 float32: 512 bytes) Using upright RootSIFT with 8000 features, nearest neighbor matching and outlier rejection enforcing local affine consistency within a confidence-based adaptive error tolerance. Matches post-processed with DEGENSAC. https://arxiv.org/abs/2006.04250 N/A 20-04-24 is_submission, is_challenge_2020
Submission ID: 00058
HardNet-Upright w/ MAGSAC (no FL...
Size: 512 bytes. Matches: built-in
7829.63 707.86
Rank: 16/127
0.486
Rank: 44/127
0.853
Rank: 32/127
0.56374
(±0.00069)
Rank: 24/127
509.07
Rank: 54/127
4250.40
Rank: 49/127
4.548
Rank: 36/127
0.412
Rank: 36/127
0.72309
(±0.00141)
Rank: 32/127
0.64341
Rank: 29/127
Challenge organizers (contact) sift8k hardnet-upright (128 float32: 512 bytes) Upright HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00122
CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 528.67
Rank: 45/127
0.472
Rank: 78/127
0.838
Rank: 61/127
0.53291
(±0.00047)
Rank: 47/127
356.56
Rank: 97/127
3550.63
Rank: 82/127
4.275
Rank: 85/127
0.443
Rank: 74/127
0.68875
(±0.00138)
Rank: 69/127
0.61083
Rank: 56/127
Challenge organizers (contact) sift8k hardnetamos (128 float32: 512 bytes) CV-DoG-HardNetAmos with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00061
L2-Net-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 358.37
Rank: 92/127
0.486
Rank: 44/127
0.845
Rank: 49/127
0.52012
(±0.00106)
Rank: 60/127
369.60
Rank: 95/127
3538.91
Rank: 85/127
4.407
Rank: 57/127
0.447
Rank: 81/127
0.68108
(±0.00177)
Rank: 77/127
0.60060
Rank: 67/127
Challenge organizers (contact) sift8k l2net-upright (128 float32: 512 bytes) Upright L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-23 is_baseline
Submission ID: 00121
CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 398.65
Rank: 78/127
0.472
Rank: 78/127
0.863
Rank: 27/127
0.53852
(±0.00039)
Rank: 42/127
356.56
Rank: 97/127
3550.63
Rank: 82/127
4.275
Rank: 85/127
0.439
Rank: 66/127
0.68875
(±0.00138)
Rank: 69/127
0.61364
Rank: 50/127
Challenge organizers (contact) sift8k hardnetamos (128 float32: 512 bytes) CV-DoG-HardNetAmos with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00069
GeoDesc w/ DEGENSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 348.52
Rank: 96/127
0.472
Rank: 86/127
0.856
Rank: 29/127
0.51112
(±0.00070)
Rank: 68/127
395.12
Rank: 90/127
3838.97
Rank: 71/127
4.264
Rank: 88/127
0.443
Rank: 72/127
0.68032
(±0.00099)
Rank: 80/127
0.59572
Rank: 73/127
Challenge organizers (contact) sift8k geodesc (128 float32: 512 bytes) GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00625
HardNet64-train-all-l2-val-14000...
Size: 512 bytes. Matches: built-in
7830.09 605.12
Rank: 29/127
0.486
Rank: 32/127
0.872
Rank: 14/127
0.58776
(±0.00048)
Rank: 7/127
899.14
Rank: 14/127
6095.78
Rank: 11/127
4.645
Rank: 11/127
0.389
Rank: 19/127
0.74849
(±0.00117)
Rank: 17/127
0.66812
Rank: 11/127
Ximin Zheng, Sheng He, Guanlin Liang (contact) sift8k hardnet64-train-all-l2-val-14000 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 14000), FLANN disabled N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00591
Guided-Hardnet-Epoch2
Size: 512 bytes. Matches: custom
7829.63 761.99
Rank: 13/127
0.486
Rank: 44/127
0.823
Rank: 88/127
0.61081
(±0.00000)
Rank: 1/127
785.57
Rank: 23/127
6330.70
Rank: 7/127
4.680
Rank: 7/127
0.358
Rank: 2/127
0.78288
(±0.00094)
Rank: 2/127
0.69684
Rank: 1/127
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet-epoch2 (128 float32: 512 bytes) sift and hardnet with 8k features, using the modified guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-24 is_submission, is_challenge_2020
Submission ID: 00086
LogPolar-Upright w/ MAGSAC (no F...
Size: 512 bytes. Matches: built-in
7829.63 732.28
Rank: 15/127
0.486
Rank: 44/127
0.844
Rank: 54/127
0.54060
(±0.00047)
Rank: 40/127
505.37
Rank: 59/127
4414.11
Rank: 43/127
4.518
Rank: 43/127
0.421
Rank: 48/127
0.71092
(±0.00251)
Rank: 43/127
0.62576
Rank: 43/127
Challenge organizers (contact) sift8k logpolar96-fixed-upright (128 float32: 512 bytes) Upright LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00106
VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 326.20
Rank: 102/127
0.490
Rank: 20/127
0.809
Rank: 97/127
0.46326
(±0.00089)
Rank: 94/127
324.62
Rank: 116/127
3030.67
Rank: 108/127
4.173
Rank: 101/127
0.462
Rank: 95/127
0.62829
(±0.00112)
Rank: 103/127
0.54578
Rank: 97/127
Challenge organizers (contact) dog vlsift (128 float32: 512 bytes) VL-DoG-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00081
LogPolar w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 591.21
Rank: 33/127
0.472
Rank: 86/127
0.832
Rank: 71/127
0.52385
(±0.00113)
Rank: 55/127
415.77
Rank: 80/127
4054.60
Rank: 60/127
4.316
Rank: 80/127
0.432
Rank: 62/127
0.69284
(±0.00429)
Rank: 63/127
0.60835
Rank: 61/127
Challenge organizers (contact) sift8k logpolar96-fixed (128 float32: 512 bytes) LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00562
HardNet64-train-all-raw64-balanc...
Size: 512 bytes. Matches: built-in
7831.92 370.11
Rank: 87/127
0.486
Rank: 73/127
0.807
Rank: 98/127
0.45770
(±0.00060)
Rank: 98/127
575.43
Rank: 39/127
4756.46
Rank: 28/127
4.563
Rank: 33/127
0.435
Rank: 65/127
0.69780
(±0.00191)
Rank: 58/127
0.57775
Rank: 81/127
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-raw64-balance-l2-224000-raw64 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 224000), FLANN disabled N/A N/A 20-05-13 is_submission, is_challenge_2020
Submission ID: 00557
HarrisZ + DeepOrientation + SOSN...
Size: 512 bytes. Matches: custom
2410.18 614.16
Rank: 24/127
0.368
Rank: 118/127
0.513
Rank: 121/127
0.05482
(±0.00000)
Rank: 121/127
626.57
Rank: 30/127
1793.00
Rank: 122/127
4.314
Rank: 82/127
0.517
Rank: 113/127
0.56933
(±0.00268)
Rank: 116/127
0.31207
Rank: 118/127
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 10) [Mishkin 2019] N/A N/A 20-05-11 is_submission, is_challenge_2020
Submission ID: 00621
CV-DoG-HardNet8-PTv2
Size: 512 bytes. Matches: built-in
7830.09 583.12
Rank: 35/127
0.486
Rank: 31/127
0.880
Rank: 2/127
0.58997
(±0.00054)
Rank: 4/127
577.37
Rank: 37/127
4476.61
Rank: 39/127
4.638
Rank: 15/127
0.407
Rank: 33/127
0.73096
(±0.00227)
Rank: 24/127
0.66046
Rank: 17/127
Milan Pultar, Dmytro Mishkin, Jiri Matas (contact) sift8k h8e512pt (128 float32: 512 bytes) HardNet8 with PCA compression, batch sampling from few images N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00520
SIFT and HardNet64 train scale(5...
Size: 512 bytes. Matches: built-in
7830.09 400.19
Rank: 77/127
0.486
Rank: 32/127
0.837
Rank: 66/127
0.50589
(±0.00059)
Rank: 70/127
619.25
Rank: 32/127
4959.78
Rank: 23/127
4.384
Rank: 60/127
0.456
Rank: 89/127
0.69172
(±0.00048)
Rank: 65/127
0.59880
Rank: 69/127
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-scale5 (128 float32: 512 bytes) SIFT with 8000 keypoints(size scaled by 5), hardnet64 with 128 descriptors, FLANN disabled N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00526
ASLFeat-MS
Size: 512 bytes. Matches: built-in
7384.19 456.55
Rank: 62/127
0.578
Rank: 5/127
0.752
Rank: 110/127
0.42208
(±0.00054)
Rank: 102/127
532.83
Rank: 46/127
4002.80
Rank: 64/127
4.521
Rank: 42/127
0.447
Rank: 78/127
0.66568
(±0.00347)
Rank: 85/127
0.54388
Rank: 99/127
Zixin Luo, Jiahui Zhang (contact) aslfeat-ms aslfeat-ms (128 float32: 512 bytes) ASLFeat (joint predition of detectors and descriptors with multi-scale input) with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC. N/A N/A 20-04-27 is_submission, is_challenge_2020
Submission ID: 00546
HarrisZ + DeepOrientation + SOSN...
Size: 512 bytes. Matches: custom
2410.18 1020.86
Rank: 6/127
0.368
Rank: 118/127
0.366
Rank: 124/127
0.07780
(±0.00000)
Rank: 120/127
1047.01
Rank: 7/127
2197.09
Rank: 119/127
4.023
Rank: 115/127
0.509
Rank: 111/127
0.59222
(±0.00276)
Rank: 111/127
0.33501
Rank: 115/127
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 15, degeneracy check true) [Mishkin 2019] N/A N/A 20-05-02 is_baseline
Submission ID: 00524
SIFT-patch
Size: 512 bytes. Matches: built-in
7861.62 84.08
Rank: 127/127
0.472
Rank: 83/127
0.574
Rank: 120/127
0.22805
(±0.00076)
Rank: 117/127
81.93
Rank: 127/127
991.42
Rank: 127/127
1.934
Rank: 127/127
0.687
Rank: 124/127
0.08662
(±0.00384)
Rank: 126/127
0.15733
Rank: 126/127
feyman_priv (contact) sift8k l2net-arcface (128 float32: 512 bytes) sift8k with l2net trained on google-landmark-dataset-v1(in 1000 class) N/A N/A 20-04-26 is_submission, is_challenge_2020
Submission ID: 00082
LogPolar w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 496.36
Rank: 53/127
0.472
Rank: 86/127
0.817
Rank: 93/127
0.49129
(±0.00011)
Rank: 78/127
399.70
Rank: 85/127
4036.37
Rank: 62/127
4.340
Rank: 73/127
0.429
Rank: 60/127
0.69100
(±0.00154)
Rank: 66/127
0.59115
Rank: 76/127
Challenge organizers (contact) sift8k logpolar96-fixed (128 float32: 512 bytes) LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00116
VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 209.26
Rank: 119/127
0.577
Rank: 6/127
0.841
Rank: 59/127
0.39330
(±0.00052)
Rank: 108/127
350.69
Rank: 104/127
3327.71
Rank: 96/127
4.076
Rank: 112/127
0.490
Rank: 108/127
0.60691
(±0.00338)
Rank: 107/127
0.50011
Rank: 110/127
Challenge organizers (contact) hessian affnetvlsift (128 float32: 512 bytes) VL-HessAffNet-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00611
sift and hardnet64 train scale(1...
Size: 512 bytes. Matches: built-in
7830.09 622.13
Rank: 22/127
0.486
Rank: 32/127
0.871
Rank: 15/127
0.58870
(±0.00041)
Rank: 5/127
899.14
Rank: 14/127
6086.16
Rank: 12/127
4.647
Rank: 10/127
0.386
Rank: 15/127
0.75127
(±0.00234)
Rank: 14/127
0.66999
Rank: 8/127
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-l2-val-14000 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 14000), FLANN disabled N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00063
SOSNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 340.09
Rank: 99/127
0.472
Rank: 86/127
0.854
Rank: 31/127
0.51891
(±0.00067)
Rank: 61/127
440.21
Rank: 71/127
4178.82
Rank: 52/127
4.378
Rank: 66/127
0.419
Rank: 45/127
0.70078
(±0.00375)
Rank: 56/127
0.60984
Rank: 59/127
Challenge organizers (contact) sift8k sosnet (128 float32: 512 bytes) SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00109
VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 250.11
Rank: 114/127
0.482
Rank: 75/127
0.838
Rank: 63/127
0.46795
(±0.00075)
Rank: 90/127
311.54
Rank: 121/127
3061.54
Rank: 105/127
4.105
Rank: 109/127
0.475
Rank: 101/127
0.62964
(±0.00428)
Rank: 99/127
0.54880
Rank: 94/127
Challenge organizers (contact) dogaffine vlsift (128 float32: 512 bytes) VL-DoGAff-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00118
VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 350.03
Rank: 94/127
0.577
Rank: 6/127
0.824
Rank: 86/127
0.46263
(±0.00079)
Rank: 95/127
350.69
Rank: 104/127
3327.71
Rank: 96/127
4.076
Rank: 112/127
0.489
Rank: 107/127
0.60691
(±0.00338)
Rank: 107/127
0.53477
Rank: 101/127
Challenge organizers (contact) hessian affnetvlsift (128 float32: 512 bytes) VL-HessAffNet-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00576
Upright-Sift + X-Net-lib w/ DEGE...
Size: 512 bytes. Matches: built-in
7830.16 465.31
Rank: 60/127
0.486
Rank: 43/127
0.844
Rank: 52/127
0.53156
(±0.00027)
Rank: 48/127
720.29
Rank: 25/127
5398.84
Rank: 19/127
4.547
Rank: 38/127
0.408
Rank: 34/127
0.71966
(±0.00160)
Rank: 38/127
0.62561
Rank: 44/127
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) sift x-net-lib (128 float32: 512 bytes) N/A N/A 20-05-20 is_submission, is_challenge_2020
Submission ID: 00075
SOSNet w/ DEGENSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 424.61
Rank: 72/127
0.472
Rank: 86/127
0.868
Rank: 19/127
0.55867
(±0.00060)
Rank: 26/127
508.66
Rank: 56/127
4502.19
Rank: 36/127
4.377
Rank: 68/127
0.404
Rank: 29/127
0.71820
(±0.00298)
Rank: 39/127
0.63844
Rank: 32/127
Challenge organizers (contact) sift8k sosnet (128 float32: 512 bytes) SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-23 is_baseline
Submission ID: 00105
VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 261.60
Rank: 113/127
0.490
Rank: 20/127
0.826
Rank: 83/127
0.46555
(±0.00055)
Rank: 93/127
324.62
Rank: 116/127
3030.67
Rank: 108/127
4.173
Rank: 101/127
0.461
Rank: 93/127
0.62829
(±0.00112)
Rank: 103/127
0.54692
Rank: 96/127
Challenge organizers (contact) dog vlsift (128 float32: 512 bytes) VL-DoG-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00062
L2-Net-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 471.61
Rank: 58/127
0.486
Rank: 44/127
0.822
Rank: 89/127
0.51450
(±0.00058)
Rank: 64/127
369.60
Rank: 95/127
3538.91
Rank: 85/127
4.407
Rank: 57/127
0.448
Rank: 83/127
0.68108
(±0.00177)
Rank: 77/127
0.59779
Rank: 70/127
Challenge organizers (contact) sift8k l2net-upright (128 float32: 512 bytes) Upright L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-22 is_baseline
Submission ID: 00048
GeoDesc w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 362.25
Rank: 90/127
0.472
Rank: 86/127
0.816
Rank: 94/127
0.46887
(±0.00025)
Rank: 89/127
350.43
Rank: 107/127
3601.67
Rank: 78/127
4.234
Rank: 92/127
0.447
Rank: 80/127
0.65553
(±0.00444)
Rank: 92/127
0.56220
Rank: 91/127
Challenge organizers (contact) sift8k geodesc (128 float32: 512 bytes) GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00056
HardNet-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 439.36
Rank: 67/127
0.486
Rank: 44/127
0.864
Rank: 25/127
0.54393
(±0.00027)
Rank: 38/127
480.13
Rank: 63/127
4167.34
Rank: 54/127
4.566
Rank: 31/127
0.417
Rank: 41/127
0.71542
(±0.00274)
Rank: 41/127
0.62967
Rank: 39/127
Challenge organizers (contact) sift8k hardnet-upright (128 float32: 512 bytes) Upright HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00613
HardNet64-data-aug-sort-51
Size: 512 bytes. Matches: built-in
7830.09 624.09
Rank: 21/127
0.486
Rank: 32/127
0.870
Rank: 16/127
0.58727
(±0.00076)
Rank: 8/127
964.80
Rank: 9/127
6350.68
Rank: 5/127
4.644
Rank: 14/127
0.383
Rank: 13/127
0.74952
(±0.00162)
Rank: 16/127
0.66839
Rank: 10/127
Ximin Zheng, Sheng He, Guanlin Liang (contact) sift8k hardnet64-data-aug-sort-51 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 124000 and data augument), FLANN disabled N/A N/A 20-05-31 is_submission, is_challenge_2020
Submission ID: 00598
HarrisZ (more kpt) + DeepOrienta...
Size: 512 bytes. Matches: custom
4478.94 563.62
Rank: 40/127
0.455
Rank: 111/127
0.711
Rank: 112/127
0.39767
(±0.00000)
Rank: 107/127
576.83
Rank: 38/127
2707.54
Rank: 115/127
4.381
Rank: 63/127
0.445
Rank: 77/127
0.69090
(±0.00163)
Rank: 68/127
0.54428
Rank: 98/127
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ (start scale 2) [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 0.5, degeneracy check true, confidence 0.98, max iter 500000) [Mishkin 2019] N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00049
GeoDesc-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 317.71
Rank: 104/127
0.486
Rank: 44/127
0.850
Rank: 42/127
0.48769
(±0.00064)
Rank: 83/127
394.67
Rank: 92/127
3863.42
Rank: 69/127
4.380
Rank: 64/127
0.440
Rank: 68/127
0.67682
(±0.00127)
Rank: 82/127
0.58225
Rank: 78/127
Challenge organizers (contact) sift8k geodesc-upright (128 float32: 512 bytes) Upright GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00057
HardNet-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 582.52
Rank: 36/127
0.486
Rank: 44/127
0.841
Rank: 57/127
0.53730
(±0.00082)
Rank: 44/127
480.13
Rank: 63/127
4167.34
Rank: 54/127
4.566
Rank: 31/127
0.416
Rank: 39/127
0.71542
(±0.00274)
Rank: 41/127
0.62636
Rank: 42/127
Challenge organizers (contact) sift8k hardnet-upright (128 float32: 512 bytes) Upright HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00047
GeoDesc w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 281.36
Rank: 109/127
0.472
Rank: 86/127
0.838
Rank: 64/127
0.47259
(±0.00033)
Rank: 88/127
350.43
Rank: 107/127
3601.67
Rank: 78/127
4.234
Rank: 92/127
0.447
Rank: 79/127
0.65553
(±0.00444)
Rank: 92/127
0.56406
Rank: 90/127
Challenge organizers (contact) sift8k geodesc (128 float32: 512 bytes) GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00558
SIFT + DeepOrientation + SOSNet ...
Size: 512 bytes. Matches: custom
2826.82 333.94
Rank: 100/127
0.284
Rank: 123/127
0.427
Rank: 123/127
0.01806
(±0.00000)
Rank: 125/127
341.75
Rank: 112/127
2003.56
Rank: 120/127
3.735
Rank: 119/127
0.593
Rank: 118/127
0.51048
(±0.00436)
Rank: 118/127
0.26427
Rank: 122/127
Fabio Bellavia (contact) sift deep-oriented-sosnet (128 float32: 512 bytes) SIFT (VLFeat implementation) [Lowe 2004] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 10) [Mishkin 2019] N/A N/A 20-05-11 is_submission, is_challenge_2020
Submission ID: 00079
LogPolar w/ DEGENSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 441.77
Rank: 66/127
0.472
Rank: 86/127
0.852
Rank: 34/127
0.53396
(±0.00086)
Rank: 46/127
415.77
Rank: 80/127
4054.60
Rank: 60/127
4.316
Rank: 80/127
0.432
Rank: 64/127
0.69284
(±0.00429)
Rank: 63/127
0.61340
Rank: 51/127
Challenge organizers (contact) sift8k logpolar96-fixed (128 float32: 512 bytes) LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00077
SOSNet-Upright w/ DEGENSAC (no F...
Size: 512 bytes. Matches: built-in
7829.63 508.38
Rank: 52/127
0.486
Rank: 44/127
0.877
Rank: 4/127
0.57385
(±0.00041)
Rank: 18/127
600.23
Rank: 34/127
4765.43
Rank: 26/127
4.584
Rank: 27/127
0.403
Rank: 27/127
0.73027
(±0.00284)
Rank: 25/127
0.65206
Rank: 24/127
Challenge organizers (contact) sift8k sosnet-upright (128 float32: 512 bytes) Upright SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-23 is_baseline
Submission ID: 00502
Guided matching with Upright roo...
Size: 512 bytes. Matches: custom
7829.27 491.96
Rank: 54/127
0.486
Rank: 29/127
0.828
Rank: 80/127
0.50310
(±0.01761)
Rank: 72/127
809.76
Rank: 20/127
5966.91
Rank: 13/127
4.462
Rank: 51/127
0.425
Rank: 55/127
0.71009
(±0.00414)
Rank: 45/127
0.60659
Rank: 64/127
(contact) sift upright-root-sift (128 float32: 512 bytes) In submission N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00015
Root-SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
7860.73 274.85
Rank: 111/127
0.472
Rank: 81/127
0.845
Rank: 50/127
0.48887
(±0.00011)
Rank: 82/127
437.74
Rank: 73/127
3814.80
Rank: 74/127
4.151
Rank: 104/127
0.458
Rank: 90/127
0.65067
(±0.00155)
Rank: 94/127
0.56977
Rank: 89/127
Challenge organizers (contact) sift-lowth rootsift (128 float32: 512 bytes) Root-SIFT with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00624
Guided-HardNet-OANet
Size: 512 bytes. Matches: custom
7829.63 765.34
Rank: 12/127
0.486
Rank: 44/127
0.820
Rank: 90/127
0.60261
(±0.00000)
Rank: 2/127
788.49
Rank: 21/127
6346.62
Rank: 6/127
4.682
Rank: 6/127
0.355
Rank: 1/127
0.78550
(±0.00157)
Rank: 1/127
0.69405
Rank: 2/127
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet-epoch2 (128 float32: 512 bytes) sift and hardnet with 8k features, first using the oanet trained from scratch then guided-matching, setting keypoint orientation to a constant value to increase performance. N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00542
sift and hardnet64 train scale(1...
Size: 512 bytes. Matches: built-in
7830.09 607.44
Rank: 28/127
0.486
Rank: 32/127
0.872
Rank: 12/127
0.58801
(±0.00037)
Rank: 6/127
950.74
Rank: 11/127
6286.36
Rank: 9/127
4.626
Rank: 17/127
0.390
Rank: 21/127
0.74662
(±0.00087)
Rank: 19/127
0.66732
Rank: 12/127
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-l2 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 138000), FLANN disabled N/A N/A 20-05-01 is_submission, is_challenge_2020
Submission ID: 00535
hardnetlib32P-Upright w/ DEGENSA...
Size: 512 bytes. Matches: built-in
7829.63 527.64
Rank: 46/127
0.486
Rank: 44/127
0.876
Rank: 6/127
0.57327
(±0.00078)
Rank: 20/127
506.73
Rank: 58/127
4230.98
Rank: 51/127
4.517
Rank: 45/127
0.418
Rank: 43/127
0.72171
(±0.00156)
Rank: 36/127
0.64749
Rank: 28/127
Chen Shen, Zhipeng Wang, Jun Zhang, Jingchao Zhou, Pengfei Xu (contact) sift8k-no-dups-mps hardnetlib32P (128 float32: 512 bytes) sift and hardnet with 8k features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-04-27 is_submission, is_challenge_2020
Submission ID: 00114
VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 348.95
Rank: 95/127
0.547
Rank: 12/127
0.800
Rank: 99/127
0.43347
(±0.00084)
Rank: 101/127
347.39
Rank: 109/127
3209.10
Rank: 101/127
4.126
Rank: 105/127
0.517
Rank: 112/127
0.58657
(±0.00120)
Rank: 113/127
0.51002
Rank: 109/127
Challenge organizers (contact) hessian vlsift (128 float32: 512 bytes) VL-Hess-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00563
HardNet64-train-all-SOS-812000
Size: 512 bytes. Matches: built-in
7830.09 612.51
Rank: 25/127
0.486
Rank: 32/127
0.869
Rank: 18/127
0.58664
(±0.00095)
Rank: 9/127
972.47
Rank: 8/127
6369.03
Rank: 4/127
4.612
Rank: 19/127
0.388
Rank: 18/127
0.74668
(±0.00251)
Rank: 18/127
0.66666
Rank: 13/127
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-sos-812000 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), sosnet64 with 128 descriptors(trained with sos loss and step 812000), FLANN disabled N/A N/A 20-05-13 is_submission, is_challenge_2020
Submission ID: 00607
HarrisZ (more kpt) + DeepOrienta...
Size: 512 bytes. Matches: custom
4478.94 746.02
Rank: 14/127
0.455
Rank: 111/127
0.699
Rank: 113/127
0.40725
(±0.00000)
Rank: 103/127
763.44
Rank: 24/127
2953.35
Rank: 111/127
4.414
Rank: 54/127
0.442
Rank: 69/127
0.69399
(±0.00179)
Rank: 62/127
0.55062
Rank: 93/127
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ (start scale 2) [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 0.75, degeneracy check true, confidence 0.98, max iter 500000) [Mishkin 2019] N/A N/A 20-06-03 is_submission, is_challenge_2020
Submission ID: 00600
ContextDesc Upright + Mutual Che...
Size: 512 bytes. Matches: custom
7830.09 624.55
Rank: 20/127
0.487
Rank: 25/127
0.847
Rank: 46/127
0.57344
(±0.00000)
Rank: 19/127
644.39
Rank: 28/127
5427.23
Rank: 18/127
4.700
Rank: 4/127
0.368
Rank: 5/127
0.77043
(±0.00133)
Rank: 4/127
0.67194
Rank: 6/127
Jiahui Zhang, Zixin Luo, Hongkai Chen (contact) contextdesc-upright contextdesc-upright (128 float32: 512 bytes) ContextDesc with 8000 SIFT features, using improved OANet matcher and DEGENSAC post-processing N/A N/A 20-05-29 is_submission, is_challenge_2020
Submission ID: 00055
HardNet-Upright w/ DEGENSAC (no ...
Size: 512 bytes. Matches: built-in
7829.63 527.59
Rank: 47/127
0.486
Rank: 44/127
0.876
Rank: 5/127
0.57279
(±0.00120)
Rank: 22/127
509.07
Rank: 54/127
4250.40
Rank: 49/127
4.548
Rank: 36/127
0.413
Rank: 37/127
0.72309
(±0.00141)
Rank: 32/127
0.64794
Rank: 27/127
Challenge organizers (contact) sift8k hardnet-upright (128 float32: 512 bytes) Upright HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00601
SIFT8k-HardNet64
Size: 512 bytes. Matches: built-in
6589.88 381.80
Rank: 82/127
0.467
Rank: 107/127
0.852
Rank: 35/127
0.52493
(±0.00044)
Rank: 53/127
420.36
Rank: 77/127
3586.82
Rank: 81/127
4.494
Rank: 48/127
0.429
Rank: 58/127
0.70580
(±0.00229)
Rank: 51/127
0.61536
Rank: 47/127
caoliang (contact) sift8k hardnet (128 float32: 512 bytes) SIFT up to 8000 keypoints, hardnet extract descriptors.Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-05-29 is_submission, is_challenge_2020
Submission ID: 00108
VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 171.63
Rank: 122/127
0.482
Rank: 75/127
0.848
Rank: 44/127
0.39839
(±0.00122)
Rank: 106/127
311.54
Rank: 121/127
3061.54
Rank: 105/127
4.105
Rank: 109/127
0.475
Rank: 103/127
0.62964
(±0.00428)
Rank: 99/127
0.51401
Rank: 107/127
Challenge organizers (contact) dogaffine vlsift (128 float32: 512 bytes) VL-DoGAff-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00594
CV-DoG-HardNet8-PT
Size: 512 bytes. Matches: built-in
7829.26 579.51
Rank: 37/127
0.484
Rank: 74/127
0.872
Rank: 13/127
0.58375
(±0.00046)
Rank: 11/127
573.22
Rank: 40/127
4489.23
Rank: 38/127
4.619
Rank: 18/127
0.403
Rank: 28/127
0.72432
(±0.00357)
Rank: 31/127
0.65403
Rank: 22/127
Milan Pultar, Dmytro Mishkin, Jiri Matas (contact) sift8k h8e512pt (128 float32: 512 bytes) HardNet8 with PCA compression N/A N/A 20-05-27 is_submission, is_challenge_2020
Submission ID: 00533
upright-sift8k-hardnet
Size: 512 bytes. Matches: built-in
6589.88 371.42
Rank: 86/127
0.467
Rank: 107/127
0.852
Rank: 39/127
0.52463
(±0.00070)
Rank: 54/127
404.02
Rank: 82/127
3501.62
Rank: 88/127
4.483
Rank: 49/127
0.428
Rank: 57/127
0.70123
(±0.00081)
Rank: 55/127
0.61293
Rank: 52/127
caoliang (contact) sift8k hardnet (128 float32: 512 bytes) SIFT up to 8000 keypoints, harnet extract descriptors.Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-24 is_submission, is_challenge_2020
Submission ID: 00614
ContextDesc Upright + Mutual Che...
Size: 512 bytes. Matches: custom
7830.09 647.63
Rank: 18/127
0.487
Rank: 25/127
0.846
Rank: 48/127
0.57826
(±0.00000)
Rank: 17/127
668.00
Rank: 26/127
5612.57
Rank: 16/127
4.666
Rank: 8/127
0.367
Rank: 4/127
0.77041
(±0.00298)
Rank: 5/127
0.67433
Rank: 5/127
Jiahui Zhang, Zixin Luo, Hongkai Chen (contact) contextdesc-upright contextdesc-upright (128 float32: 512 bytes) ContextDesc with 8000 SIFT features, using improved OANet matcher and DEGENSAC post-processing N/A N/A 20-05-31 is_submission, is_challenge_2020
Submission ID: 00053
HardNet w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 575.07
Rank: 39/127
0.472
Rank: 86/127
0.842
Rank: 56/127
0.55022
(±0.00001)
Rank: 34/127
426.84
Rank: 74/127
4001.40
Rank: 65/127
4.339
Rank: 75/127
0.419
Rank: 47/127
0.70962
(±0.00191)
Rank: 46/127
0.62992
Rank: 38/127
Challenge organizers (contact) sift8k hardnet (128 float32: 512 bytes) HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00005
SURF (OpenCV), DEGENSAC
Size: 256 bytes. Matches: built-in
7728.57 125.94
Rank: 126/127
0.432
Rank: 117/127
0.684
Rank: 114/127
0.24948
(±0.00065)
Rank: 116/127
928.29
Rank: 12/127
3455.28
Rank: 89/127
3.428
Rank: 122/127
0.733
Rank: 125/127
0.40490
(±0.00448)
Rank: 124/127
0.32719
Rank: 117/127
Challenge organizers (contact) surf-lowth surf (64 float32: 256 bytes) SURF with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00510
sift and hardnet64
Size: 512 bytes. Matches: built-in
7861.62 326.87
Rank: 101/127
0.472
Rank: 84/127
0.845
Rank: 51/127
0.51149
(±0.00018)
Rank: 67/127
521.58
Rank: 50/127
4587.88
Rank: 33/127
4.222
Rank: 95/127
0.424
Rank: 54/127
0.69458
(±0.00340)
Rank: 61/127
0.60304
Rank: 65/127
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-v2 (128 float32: 512 bytes) SIFT with 8000 keypoints, hardnet64 with 128 descriptors N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00559
SIFT + DeepOrientation + SOSNet ...
Size: 512 bytes. Matches: custom
2826.82 164.43
Rank: 123/127
0.284
Rank: 123/127
0.650
Rank: 116/127
0.05206
(±0.00000)
Rank: 122/127
167.59
Rank: 126/127
1549.91
Rank: 124/127
3.989
Rank: 118/127
0.622
Rank: 120/127
0.47700
(±0.00167)
Rank: 120/127
0.26453
Rank: 121/127
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) SIFT (VLFeat implementation) [Lowe 2004] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 5) [Mishkin 2019] N/A N/A 20-05-10 is_submission, is_challenge_2020
Submission ID: 00070
GeoDesc-Upright w/ MAGSAC (no FL...
Size: 512 bytes. Matches: built-in
7829.63 534.05
Rank: 44/127
0.486
Rank: 44/127
0.848
Rank: 45/127
0.52078
(±0.00028)
Rank: 59/127
458.60
Rank: 66/127
4146.83
Rank: 57/127
4.412
Rank: 55/127
0.432
Rank: 63/127
0.70442
(±0.00189)
Rank: 53/127
0.61260
Rank: 54/127
Challenge organizers (contact) sift8k geodesc-upright (128 float32: 512 bytes) Upright GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00052
HardNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 363.51
Rank: 89/127
0.472
Rank: 86/127
0.853
Rank: 33/127
0.52325
(±0.00098)
Rank: 56/127
402.15
Rank: 83/127
3895.00
Rank: 67/127
4.345
Rank: 70/127
0.426
Rank: 56/127
0.69653
(±0.00108)
Rank: 59/127
0.60989
Rank: 58/127
Challenge organizers (contact) sift8k hardnet (128 float32: 512 bytes) HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00549
SIFT + DeepOrientation + SOSNet ...
Size: 512 bytes. Matches: custom
2826.82 216.23
Rank: 118/127
0.284
Rank: 123/127
0.456
Rank: 122/127
0.14544
(±0.00000)
Rank: 119/127
221.40
Rank: 125/127
1839.87
Rank: 121/127
3.608
Rank: 120/127
0.641
Rank: 121/127
0.47073
(±0.00158)
Rank: 121/127
0.30808
Rank: 119/127
Fabio Bellavia (contact) sift deep-oriented-sosnet (128 float32: 512 bytes) SIFT (VLFeat implementation) [Lowe 2004] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 0.75, degeneracy check true) [Mishkin 2019] N/A N/A 20-05-03 is_submission, is_challenge_2020
Submission ID: 00113
VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 290.24
Rank: 108/127
0.547
Rank: 12/127
0.814
Rank: 95/127
0.44501
(±0.00077)
Rank: 100/127
347.39
Rank: 109/127
3209.10
Rank: 101/127
4.126
Rank: 105/127
0.519
Rank: 115/127
0.58657
(±0.00120)
Rank: 113/127
0.51579
Rank: 105/127
Challenge organizers (contact) hessian vlsift (128 float32: 512 bytes) VL-Hess-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00567
Guided-HardNet32-v1-lib-qht-p
Size: 512 bytes. Matches: custom
7829.63 520.40
Rank: 48/127
0.486
Rank: 44/127
0.875
Rank: 8/127
0.58509
(±0.00000)
Rank: 10/127
536.93
Rank: 45/127
4685.16
Rank: 31/127
4.645
Rank: 12/127
0.376
Rank: 9/127
0.75702
(±0.00159)
Rank: 12/127
0.67105
Rank: 7/127
Chen Shen, Zhipeng Wang, Jun Zhang, Zhiwei Ruan, Zhongkun Chen, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet32-v1-lib-qht-p (128 float32: 512 bytes) sift and sosnet with 8k features, using the guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-15 is_submission, is_challenge_2020
Submission ID: 00547
SIFT + DeepOrientation + SOSNet ...
Size: 512 bytes. Matches: custom
2826.82 1078.85
Rank: 4/127
0.284
Rank: 123/127
0.166
Rank: 126/127
0.04936
(±0.00000)
Rank: 123/127
1106.68
Rank: 5/127
2494.00
Rank: 117/127
3.400
Rank: 123/127
0.646
Rank: 122/127
0.43054
(±0.00185)
Rank: 122/127
0.23995
Rank: 124/127
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) SIFT (VLFeat implementation) [Lowe 2004] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 15, degeneracy check true) [Mishkin 2019] N/A N/A 20-05-03 is_submission, is_challenge_2020
Submission ID: 00537
SIFT + DeepOrientation + SOSNet ...
Size: 512 bytes. Matches: custom
2826.82 1476.71
Rank: 1/127
0.284
Rank: 123/127
0.145
Rank: 127/127
0.00178
(±0.00000)
Rank: 127/127
1506.82
Rank: 2/127
2538.44
Rank: 116/127
3.388
Rank: 125/127
0.669
Rank: 123/127
0.42682
(±0.00259)
Rank: 123/127
0.21430
Rank: 125/127
Fabio Bellavia (contact) sift deep-oriented-sosnet (128 float32: 512 bytes) SIFT (VLFeat implementation) [Lowe 2004] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] N/A N/A 20-05-01 is_submission, is_challenge_2020
Submission ID: 00110
VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 317.13
Rank: 105/127
0.482
Rank: 75/127
0.820
Rank: 91/127
0.46657
(±0.00079)
Rank: 92/127
311.54
Rank: 121/127
3061.54
Rank: 105/127
4.105
Rank: 109/127
0.475
Rank: 102/127
0.62964
(±0.00428)
Rank: 99/127
0.54810
Rank: 95/127
Challenge organizers (contact) dogaffine vlsift (128 float32: 512 bytes) VL-DoGAff-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00072
L2-Net-Upright w/ DEGENSAC (no F...
Size: 512 bytes. Matches: built-in
7829.63 435.68
Rank: 68/127
0.486
Rank: 44/127
0.859
Rank: 28/127
0.54497
(±0.00003)
Rank: 36/127
395.53
Rank: 88/127
3603.85
Rank: 76/127
4.382
Rank: 61/127
0.452
Rank: 86/127
0.68491
(±0.00338)
Rank: 74/127
0.61494
Rank: 49/127
Challenge organizers (contact) sift8k l2net-upright (128 float32: 512 bytes) Upright L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-22 is_baseline
Submission ID: 00590
Guided-HardNet-epoch4
Size: 512 bytes. Matches: custom
7829.63 586.24
Rank: 34/127
0.486
Rank: 44/127
0.875
Rank: 7/127
0.59919
(±0.00000)
Rank: 3/127
604.81
Rank: 33/127
5062.27
Rank: 22/127
4.710
Rank: 3/127
0.369
Rank: 7/127
0.76219
(±0.00253)
Rank: 7/127
0.68069
Rank: 3/127
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet-epoch4 (128 float32: 512 bytes) sift and hardnet with 8k features, using the modified guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-24 is_submission, is_challenge_2020
Submission ID: 00117
VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 299.02
Rank: 106/127
0.577
Rank: 6/127
0.831
Rank: 72/127
0.46793
(±0.00035)
Rank: 91/127
350.69
Rank: 104/127
3327.71
Rank: 96/127
4.076
Rank: 112/127
0.487
Rank: 105/127
0.60691
(±0.00338)
Rank: 107/127
0.53742
Rank: 100/127
Challenge organizers (contact) hessian affnetvlsift (128 float32: 512 bytes) VL-HessAffNet-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00610
Hardnet-Upright-AdaLAM
Size: 512 bytes. Matches: custom
6556.61 627.71
Rank: 19/127
0.442
Rank: 114/127
0.828
Rank: 79/127
0.58300
(±0.00000)
Rank: 12/127
645.47
Rank: 27/127
5074.91
Rank: 21/127
4.575
Rank: 30/127
0.361
Rank: 3/127
0.77056
(±0.00064)
Rank: 3/127
0.67678
Rank: 4/127
Luca Cavalli, Viktor Larsson, Martin Oswald, Torsten Sattler, Marc Pollefeys (contact) sift-def hardnet-upright (128 float32: 512 bytes) Using upright Hardnet descriptors with 8000 features, nearest neighbor matching and outlier rejection enforcing local affine consistency within a confidence-based adaptive error tolerance. Matches post-processed with DEGENSAC. https://arxiv.org/abs/2006.04250 N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00074
L2-Net w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 481.03
Rank: 55/127
0.472
Rank: 86/127
0.830
Rank: 78/127
0.52519
(±0.00039)
Rank: 51/127
339.00
Rank: 113/127
3424.94
Rank: 90/127
4.206
Rank: 97/127
0.448
Rank: 82/127
0.66445
(±0.00213)
Rank: 87/127
0.59482
Rank: 74/127
Challenge organizers (contact) sift8k l2net (128 float32: 512 bytes) L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-23 is_baseline
Submission ID: 00595
SIFT8k_8000_HardNet64-train-all-...
Size: 512 bytes. Matches: built-in
7830.09 600.42
Rank: 31/127
0.486
Rank: 32/127
0.867
Rank: 20/127
0.58265
(±0.00102)
Rank: 14/127
786.40
Rank: 22/127
5595.19
Rank: 17/127
4.603
Rank: 22/127
0.390
Rank: 20/127
0.74372
(±0.00317)
Rank: 23/127
0.66319
Rank: 16/127
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-ben86-l2-210000 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 210000 and iou 0.86), FLANN disabled N/A N/A 20-05-27 is_submission, is_challenge_2020
Submission ID: 00618
Sift-Sem + HyNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7530.17 476.84
Rank: 57/127
0.490
Rank: 19/127
0.882
Rank: 1/127
0.57281
(±0.00047)
Rank: 21/127
809.82
Rank: 19/127
5868.33
Rank: 14/127
4.585
Rank: 26/127
0.390
Rank: 22/127
0.74557
(±0.00393)
Rank: 21/127
0.65919
Rank: 19/127
Barroso-Laguna, Axel and Tian, Yurun, Ng, Tony and Mikolajczyk, Krystian (contact) sift-semantics hynet (128 float32: 512 bytes) N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00521
SIFT8k-gift
Size: 512 bytes. Matches: built-in
6589.48 426.77
Rank: 71/127
0.467
Rank: 110/127
0.830
Rank: 74/127
0.48128
(±0.00056)
Rank: 86/127
452.18
Rank: 68/127
3770.17
Rank: 75/127
4.595
Rank: 25/127
0.444
Rank: 76/127
0.67443
(±0.00253)
Rank: 84/127
0.57785
Rank: 80/127
Chen Shen (contact) sift-def gift (128 float32: 512 bytes) sift-nodups, gift with 8k features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-04-26 is_submission, is_challenge_2020
Submission ID: 00566
Guided-HardNet2qhtp.json
Size: 512 bytes. Matches: custom
7829.63 517.72
Rank: 49/127
0.486
Rank: 44/127
0.874
Rank: 9/127
0.58251
(±0.00000)
Rank: 15/127
448.12
Rank: 70/127
4072.82
Rank: 59/127
4.607
Rank: 20/127
0.387
Rank: 17/127
0.74640
(±0.00246)
Rank: 20/127
0.66445
Rank: 14/127
Chen Shen, Zhipeng Wang, Jun Zhang, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnetnd32qhtp (128 float32: 512 bytes) sift and hardnet with 8k features, using the guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-15 is_submission, is_challenge_2020
Submission ID: 00572
Guided HardNet qht
Size: 512 bytes. Matches: custom
7829.63 885.09
Rank: 8/127
0.486
Rank: 44/127
0.774
Rank: 106/127
0.54663
(±0.00000)
Rank: 35/127
538.28
Rank: 43/127
4687.01
Rank: 30/127
4.644
Rank: 13/127
0.376
Rank: 10/127
0.75721
(±0.00140)
Rank: 11/127
0.65192
Rank: 25/127
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet-qht (128 float32: 512 bytes) sift and hardnet with 8k features, using the guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-20 is_submission, is_challenge_2020
Submission ID: 00541
sift and sosnet64 train scale(12...
Size: 512 bytes. Matches: built-in
7830.09 603.56
Rank: 30/127
0.486
Rank: 32/127
0.866
Rank: 21/127
0.58268
(±0.00061)
Rank: 13/127
957.95
Rank: 10/127
6312.38
Rank: 8/127
4.607
Rank: 21/127
0.385
Rank: 14/127
0.74476
(±0.00137)
Rank: 22/127
0.66372
Rank: 15/127
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-sos (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), sosnet64 with 128 descriptors(trained with sos loss and step 348000), FLANN disabled N/A N/A 20-05-01 is_submission, is_challenge_2020
Submission ID: 00102
KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 815.40
Rank: 10/127
0.582
Rank: 2/127
0.788
Rank: 101/127
0.47395
(±0.00073)
Rank: 87/127
356.21
Rank: 100/127
3366.01
Rank: 93/127
4.319
Rank: 77/127
0.464
Rank: 96/127
0.64829
(±0.00174)
Rank: 95/127
0.56112
Rank: 92/127
Challenge organizers (contact) keynettuned vlhardnet (128 float32: 512 bytes) KeyNet-HardNet with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00556
HarrisZ + DeepOrientation + SOSN...
Size: 512 bytes. Matches: custom
2410.18 385.16
Rank: 81/127
0.368
Rank: 118/127
0.646
Rank: 118/127
0.04192
(±0.00000)
Rank: 124/127
392.52
Rank: 94/127
1475.19
Rank: 125/127
4.454
Rank: 52/127
0.601
Rank: 119/127
0.48498
(±0.00118)
Rank: 119/127
0.26345
Rank: 123/127
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 5) [Mishkin 2019] N/A N/A 20-05-11 is_submission, is_challenge_2020
Submission ID: 00073
L2-Net-Upright w/ MAGSAC (no FLA...
Size: 512 bytes. Matches: built-in
7829.63 577.91
Rank: 38/127
0.486
Rank: 44/127
0.837
Rank: 67/127
0.53911
(±0.00062)
Rank: 41/127
395.53
Rank: 88/127
3603.85
Rank: 76/127
4.382
Rank: 61/127
0.455
Rank: 88/127
0.68491
(±0.00338)
Rank: 74/127
0.61201
Rank: 55/127
Challenge organizers (contact) sift8k l2net-upright (128 float32: 512 bytes) Upright L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-22 is_baseline
Submission ID: 00080
LogPolar w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 376.42
Rank: 84/127
0.472
Rank: 86/127
0.837
Rank: 65/127
0.50188
(±0.00044)
Rank: 73/127
399.70
Rank: 85/127
4036.37
Rank: 62/127
4.340
Rank: 73/127
0.429
Rank: 59/127
0.69100
(±0.00154)
Rank: 66/127
0.59644
Rank: 72/127
Challenge organizers (contact) sift8k logpolar96-fixed (128 float32: 512 bytes) LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00006
Upright SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
7829.24 319.29
Rank: 103/127
0.487
Rank: 23/127
0.830
Rank: 76/127
0.48742
(±0.00042)
Rank: 84/127
525.57
Rank: 47/127
4147.20
Rank: 56/127
4.258
Rank: 90/127
0.460
Rank: 92/127
0.65664
(±0.00231)
Rank: 91/127
0.57203
Rank: 87/127
Challenge organizers (contact) sift-lowth sift-upright (128 float32: 512 bytes) Upright SIFT with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00059
L2-Net w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 297.34
Rank: 107/127
0.472
Rank: 86/127
0.835
Rank: 70/127
0.49523
(±0.00042)
Rank: 77/127
314.54
Rank: 119/127
3314.19
Rank: 99/127
4.204
Rank: 99/127
0.467
Rank: 99/127
0.65665
(±0.00287)
Rank: 89/127
0.57594
Rank: 83/127
Challenge organizers (contact) sift8k l2net (128 float32: 512 bytes) L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-22 is_baseline
Submission ID: 00050
GeoDesc-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 410.23
Rank: 74/127
0.486
Rank: 44/127
0.827
Rank: 81/127
0.48259
(±0.00060)
Rank: 85/127
394.67
Rank: 92/127
3863.42
Rank: 69/127
4.380
Rank: 64/127
0.443
Rank: 73/127
0.67682
(±0.00127)
Rank: 82/127
0.57971
Rank: 79/127
Challenge organizers (contact) sift8k geodesc-upright (128 float32: 512 bytes) Upright GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00054
HardNet w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 477.64
Rank: 56/127
0.472
Rank: 86/127
0.830
Rank: 77/127
0.51809
(±0.00053)
Rank: 62/127
402.15
Rank: 83/127
3895.00
Rank: 67/127
4.345
Rank: 70/127
0.423
Rank: 53/127
0.69653
(±0.00108)
Rank: 59/127
0.60731
Rank: 63/127
Challenge organizers (contact) sift8k hardnet (128 float32: 512 bytes) HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00067
GeoDesc w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 453.39
Rank: 63/127
0.472
Rank: 86/127
0.835
Rank: 68/127
0.50559
(±0.00081)
Rank: 71/127
395.12
Rank: 90/127
3838.97
Rank: 71/127
4.264
Rank: 88/127
0.442
Rank: 71/127
0.68032
(±0.00099)
Rank: 80/127
0.59295
Rank: 75/127
Challenge organizers (contact) sift8k geodesc (128 float32: 512 bytes) GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00540
ASLV2+OANetV2+DEGENSAC
Size: 512 bytes. Matches: custom
5982.87 1044.40
Rank: 5/127
0.550
Rank: 11/127
0.758
Rank: 108/127
0.50173
(±0.00000)
Rank: 74/127
1067.11
Rank: 6/127
4587.47
Rank: 34/127
4.948
Rank: 2/127
0.386
Rank: 16/127
0.75275
(±0.00186)
Rank: 13/127
0.62724
Rank: 41/127
Jiahui Zhang, Zixin Luo, Hongkai Chen (contact) aslv2 aslv2 (128 float32: 512 bytes) ASL detector and descriptor, 8k keypoints, using improved OANet matcher and DEGENSAC post-processing N/A N/A 20-05-01 is_submission, is_challenge_2020
Submission ID: 00514
UprightRootSIFT-AdaLAM
Size: 512 bytes. Matches: custom
6449.42 619.87
Rank: 23/127
0.436
Rank: 115/127
0.758
Rank: 109/127
0.31372
(±0.00000)
Rank: 113/127
640.08
Rank: 29/127
5286.27
Rank: 20/127
4.343
Rank: 72/127
0.398
Rank: 25/127
0.72130
(±0.00085)
Rank: 37/127
0.51751
Rank: 104/127
Luca Cavalli, Viktor Larsson, Martin Oswald, Torsten Sattler, Marc Pollefeys (contact) sift-def rootsift-upright (128 float32: 512 bytes) Using upright RootSIFT with 8000 features, nearest neighbor matching and outlier rejection enforcing local affine consistency within a confidence-based adaptive error tolerance. https://arxiv.org/abs/2006.04250 N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00511
r2d2-wasfi
Size: 512 bytes. Matches: built-in
7861.03 276.55
Rank: 110/127
0.503
Rank: 17/127
0.762
Rank: 107/127
0.37893
(±0.00023)
Rank: 109/127
416.35
Rank: 79/127
2834.69
Rank: 112/127
4.293
Rank: 84/127
0.480
Rank: 104/127
0.64359
(±0.00103)
Rank: 98/127
0.51126
Rank: 108/127
Chen Shen (contact) r2d2 r2d2-wasfi-epoch25-pretrained-n16-8k (128 float32: 512 bytes) r2d2 with 8k features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00083
LogPolar-Upright w/ DEGENSAC (no...
Size: 512 bytes. Matches: built-in
7829.63 543.18
Rank: 42/127
0.486
Rank: 44/127
0.865
Rank: 24/127
0.55102
(±0.00015)
Rank: 33/127
505.37
Rank: 59/127
4414.11
Rank: 43/127
4.518
Rank: 43/127
0.422
Rank: 50/127
0.71092
(±0.00251)
Rank: 43/127
0.63097
Rank: 37/127
Challenge organizers (contact) sift8k logpolar96-fixed-upright (128 float32: 512 bytes) Upright LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00064
SOSNet w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 444.55
Rank: 65/127
0.472
Rank: 86/127
0.831
Rank: 73/127
0.51402
(±0.00057)
Rank: 65/127
440.21
Rank: 71/127
4178.82
Rank: 52/127
4.378
Rank: 66/127
0.422
Rank: 51/127
0.70078
(±0.00375)
Rank: 56/127
0.60740
Rank: 62/127
Challenge organizers (contact) sift8k sosnet (128 float32: 512 bytes) SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00602
retrained sosnet
Size: 512 bytes. Matches: built-in
7861.11 340.97
Rank: 98/127
0.472
Rank: 86/127
0.840
Rank: 60/127
0.48990
(±0.00046)
Rank: 81/127
623.55
Rank: 31/127
4812.36
Rank: 25/127
4.233
Rank: 94/127
0.458
Rank: 91/127
0.66502
(±0.00158)
Rank: 86/127
0.57746
Rank: 82/127
sosnet (contact) siftdef sosnet (128 float32: 512 bytes) default setting and retrained sosnet N/A N/A 20-05-30 is_submission, is_challenge_2020
Submission ID: 00104
VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 179.68
Rank: 121/127
0.490
Rank: 20/127
0.835
Rank: 69/127
0.39987
(±0.00068)
Rank: 104/127
324.62
Rank: 116/127
3030.67
Rank: 108/127
4.173
Rank: 101/127
0.467
Rank: 100/127
0.62829
(±0.00112)
Rank: 103/127
0.51408
Rank: 106/127
Challenge organizers (contact) dog vlsift (128 float32: 512 bytes) VL-DoG-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline

Phototourism: restricted keypoints, standard descriptors (512 bytes)

Note: entries with the same multi-view configuration may seem duplicated. This is normal: performance is averaged across tasks.
Stereo Multiview Avg.
Method NF NI Rep.
(3 pix.)
MS
(3 pix.)
mAA
(at 10o)
NM NL TL ATE mAA
(at 100)
mAA
(at 100)
Authors Keypoint Descriptor Summary Paper Website Processing date Flags
Submission ID: 00555
SuperPoint+GIFT+Graph Motion Coh...
Size: 512 bytes. Matches: custom
1940.85 281.06
Rank: 8/77
0.356
Rank: 22/77
0.676
Rank: 66/77
0.42283
(±0.00000)
Rank: 32/77
290.06
Rank: 17/77
1887.70
Rank: 12/77
4.567
Rank: 14/77
0.420
Rank: 7/77
0.70395
(±0.00239)
Rank: 6/77
0.56339
Rank: 15/77
Yuan Liu (contact) superpoint-2k scale-gift (128 float32: 512 bytes) Detecting by SuperPoint, descriptors from GIFT, matching by Graph Motion Coherence Network, geometry estimated by DEGENSAC with inlier threshold 1.0 N/A N/A 20-05-10 is_submission, is_challenge_2020
Submission ID: 00038
SOSNet, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 134.59
Rank: 41/77
0.323
Rank: 58/77
0.760
Rank: 55/77
0.38654
(±0.00056)
Rank: 48/77
145.47
Rank: 67/77
1271.41
Rank: 52/77
4.045
Rank: 57/77
0.526
Rank: 54/77
0.56072
(±0.00077)
Rank: 54/77
0.47363
Rank: 51/77
Challenge organizers (contact) sift-def sosnet (128 float32: 512 bytes) SOSNet descriptors extracted on SIFT keypoints and MAGSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00705
Upright SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.72 104.74
Rank: 61/77
0.333
Rank: 34/77
0.764
Rank: 53/77
0.37292
(±0.00081)
Rank: 55/77
204.13
Rank: 34/77
1293.75
Rank: 49/77
4.008
Rank: 61/77
0.567
Rank: 66/77
0.52808
(±0.00416)
Rank: 61/77
0.45050
Rank: 59/77
Challenge organizers (contact) sift-def sift-upright (128 float32: 512 bytes) Upright SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-06-01 is_baseline
Submission ID: 00623
HardNet-epoch2-OANet
Size: 512 bytes. Matches: custom
2047.76 247.45
Rank: 11/77
0.347
Rank: 24/77
0.765
Rank: 51/77
0.44717
(±0.00000)
Rank: 20/77
254.11
Rank: 24/77
1898.45
Rank: 11/77
4.503
Rank: 17/77
0.437
Rank: 13/77
0.67958
(±0.00183)
Rank: 13/77
0.56337
Rank: 16/77
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift2k hardnet-epoch2 (128 float32: 512 bytes) sift and hardnet with 2k features, using the oanet trained from scratch and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00007
L2-Net, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 87.04
Rank: 68/77
0.323
Rank: 64/77
0.779
Rank: 40/77
0.36265
(±0.00090)
Rank: 57/77
138.97
Rank: 69/77
1211.49
Rank: 58/77
3.948
Rank: 65/77
0.537
Rank: 59/77
0.52019
(±0.00120)
Rank: 63/77
0.44142
Rank: 61/77
Challenge organizers (contact) sift-def l2net (128 float32: 512 bytes) L2-Net descriptors extracted on SIFT keypoints and DEGENSAC for stereo. Please refer to the baselines repository (linked) for details. http://www.nlpr.ia.ac.cn/fanbin/pub/L2-Net_CVPR17.pdf https://github.com/vcg-uvic/image-matching-benchmark-baselines 20-04-23 is_baseline
Submission ID: 00606
HardNet-epoch2-OANet
Size: 512 bytes. Matches: custom
2047.76 244.62
Rank: 14/77
0.347
Rank: 24/77
0.773
Rank: 46/77
0.46389
(±0.00000)
Rank: 11/77
251.11
Rank: 26/77
1916.44
Rank: 10/77
4.501
Rank: 18/77
0.428
Rank: 11/77
0.69407
(±0.00216)
Rank: 10/77
0.57898
Rank: 8/77
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift2k hardnet-epoch2 (128 float32: 512 bytes) sift and hardnet with 2k features, using the oanet and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-30 is_submission, is_challenge_2020
Submission ID: 00604
Key.Net+GIFT+GMCNet+DEGENSAC
Size: 512 bytes. Matches: custom
2046.07 377.20
Rank: 4/77
0.450
Rank: 4/77
0.776
Rank: 42/77
0.45282
(±0.00000)
Rank: 16/77
386.63
Rank: 11/77
1474.77
Rank: 27/77
4.663
Rank: 8/77
0.420
Rank: 6/77
0.70500
(±0.00294)
Rank: 5/77
0.57891
Rank: 9/77
Yuan Liu (contact) keynet-2k scale-gift (128 float32: 512 bytes) Detecting by Key.Net, descriptors from GIFT, matching by Graph Motion Coherence Network, geometry estimated by DEGENSAC with inlier threshold 0.7 N/A N/A 20-05-30 is_submission, is_challenge_2020
Submission ID: 00703
Upright Root-SIFT (OpenCV), DEGE...
Size: 512 bytes. Matches: built-in
1892.72 112.27
Rank: 55/77
0.333
Rank: 34/77
0.782
Rank: 37/77
0.39860
(±0.00077)
Rank: 42/77
199.34
Rank: 37/77
1341.66
Rank: 41/77
4.090
Rank: 53/77
0.518
Rank: 49/77
0.56230
(±0.00234)
Rank: 52/77
0.48045
Rank: 46/77
Challenge organizers (contact) sift-def rootsift-upright (128 float32: 512 bytes) Upright Root-SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-06-01 is_baseline
Submission ID: 00516
Key.Net (MS 4-1, Pyramid) + X-Ne...
Size: 512 bytes. Matches: built-in
2039.11 265.52
Rank: 10/77
0.447
Rank: 9/77
0.776
Rank: 44/77
0.40559
(±0.00112)
Rank: 37/77
336.81
Rank: 12/77
1559.74
Rank: 22/77
4.536
Rank: 15/77
0.461
Rank: 24/77
0.66406
(±0.00521)
Rank: 19/77
0.53483
Rank: 28/77
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) keynet-41-pyramidv2 x-net-lib (128 float32: 512 bytes) N/A N/A 20-04-24 is_submission, is_challenge_2020
Submission ID: 00002
SURF (OpenCV), DEGENSAC
Size: 256 bytes. Matches: built-in
2010.88 71.90
Rank: 71/77
0.321
Rank: 70/77
0.634
Rank: 67/77
0.20761
(±0.00029)
Rank: 70/77
300.02
Rank: 16/77
1208.06
Rank: 61/77
3.549
Rank: 70/77
0.695
Rank: 70/77
0.37845
(±0.00399)
Rank: 70/77
0.29303
Rank: 70/77
Challenge organizers (contact) surf-def surf (64 float32: 256 bytes) SURF with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00544
Key.Net + X-Net(Lib) w/ DEGENSAC
Size: 512 bytes. Matches: built-in
2039.11 194.10
Rank: 16/77
0.447
Rank: 9/77
0.814
Rank: 12/77
0.42332
(±0.00096)
Rank: 31/77
312.61
Rank: 14/77
1515.67
Rank: 24/77
4.567
Rank: 13/77
0.448
Rank: 16/77
0.66559
(±0.00233)
Rank: 18/77
0.54445
Rank: 21/77
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) keynet x-net-lib (128 float32: 512 bytes) N/A N/A 20-04-28 is_submission, is_challenge_2020
Submission ID: 00523
SEKD
Size: 512 bytes. Matches: built-in
2043.44 116.81
Rank: 49/77
0.386
Rank: 17/77
0.799
Rank: 26/77
0.42811
(±0.00096)
Rank: 29/77
166.41
Rank: 53/77
1163.31
Rank: 64/77
4.427
Rank: 22/77
0.456
Rank: 21/77
0.64919
(±0.00171)
Rank: 22/77
0.53865
Rank: 25/77
Yafei Song, Ling Cai, Mingyang Li (contact) sekd sekd (128 float32: 512 bytes) Our method named SEKD: Self-Evolving Keypoint Detection and Description, where the SEKD model is trained using COCO test images. In this submission each image has up to 2048 SEKD keypoints, and 128-dim float descriptor. We use the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC. N/A N/A 20-04-29 is_submission, is_challenge_2020
Submission ID: 00575
SIFT2k-2048-HardNet64-rain-all-s...
Size: 512 bytes. Matches: built-in
2047.77 161.77
Rank: 24/77
0.339
Rank: 30/77
0.814
Rank: 10/77
0.43543
(±0.00051)
Rank: 25/77
255.45
Rank: 22/77
1708.29
Rank: 17/77
4.327
Rank: 32/77
0.451
Rank: 18/77
0.64653
(±0.00221)
Rank: 23/77
0.54098
Rank: 23/77
Ximin Zheng, Sheng He, Hualong Shi (contact) sift2k sift2k-2048-hardnet64-train-all-sos-812000 (128 float32: 512 bytes) SIFT with 2048 keypoints(scale 12), sosnet64 with 128 descriptors(trained with sos loss and step 812000), FLANN disabled N/A N/A 20-05-20 is_submission, is_challenge_2020
Submission ID: 00599
SEKD
Size: 512 bytes. Matches: built-in
2043.44 129.46
Rank: 45/77
0.386
Rank: 17/77
0.809
Rank: 16/77
0.45066
(±0.00057)
Rank: 17/77
176.60
Rank: 46/77
1209.63
Rank: 60/77
4.437
Rank: 21/77
0.454
Rank: 19/77
0.66094
(±0.00227)
Rank: 20/77
0.55580
Rank: 17/77
Yafei Song, Ling Cai, Jia Li, Yonghong Tian, Mingyang Li (contact) sekd sekd (128 float32: 512 bytes) Our method named SEKD: Self-Evolving Keypoint Detection and Description, where the SEKD model is trained using COCO validation set. In this submission each image has up to 2048 SEKD keypoints, and 128-dim float descriptor. We use the built-in matcher (bidirectional filter with the both strategy, without flann, optimal inlier and ratio test thresholds) with DEGENSAC. N/A N/A 20-05-29 is_submission, is_challenge_2020
Submission ID: 00001
HardNet, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 109.52
Rank: 58/77
0.323
Rank: 58/77
0.776
Rank: 43/77
0.38578
(±0.00131)
Rank: 50/77
153.29
Rank: 61/77
1306.64
Rank: 46/77
4.026
Rank: 59/77
0.539
Rank: 62/77
0.55573
(±0.00175)
Rank: 58/77
0.47075
Rank: 53/77
Challenge organizers (contact) sift-def hardnet (128 float32: 512 bytes) HardNet descriptors extracted on SIFT keypoints and DEGENSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00565
SIFT2k_2000_HardNet64-train-all-...
Size: 512 bytes. Matches: built-in
1999.83 157.70
Rank: 26/77
0.336
Rank: 32/77
0.813
Rank: 13/77
0.43188
(±0.00005)
Rank: 26/77
248.91
Rank: 27/77
1666.27
Rank: 19/77
4.323
Rank: 33/77
0.465
Rank: 27/77
0.64404
(±0.00271)
Rank: 25/77
0.53796
Rank: 26/77
Ximin Zheng, Sheng He, Hualong Shi (contact) sift2k sift2k-2000-hardnet64-train-all-sos-812000 (128 float32: 512 bytes) SIFT with 2000 keypoints(scale 12), sosnet64 with 128 descriptors(trained with sos loss and step 812000), FLANN disabled N/A N/A 20-05-13 is_submission, is_challenge_2020
Submission ID: 00620
pffNet + SuperPoint + MAGSAC
Size: 512 bytes. Matches: built-in
1267.22 57.48
Rank: 75/77
0.341
Rank: 28/77
0.632
Rank: 68/77
0.22549
(±0.00077)
Rank: 68/77
131.91
Rank: 71/77
893.88
Rank: 74/77
4.339
Rank: 28/77
0.531
Rank: 57/77
0.55935
(±0.00143)
Rank: 56/77
0.39242
Rank: 67/77
Jongmin Lee, Seungwook Kim, Yoonwoo Jeong (contact) superpoint pffnet (128 float32: 512 bytes) pffNet descriptors + SuperPoint keypoints + MAGSAC outlier-filtering N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00011
GeoDesc, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 93.44
Rank: 66/77
0.323
Rank: 64/77
0.759
Rank: 56/77
0.34215
(±0.00054)
Rank: 60/77
120.68
Rank: 74/77
1098.70
Rank: 68/77
4.002
Rank: 62/77
0.552
Rank: 64/77
0.50594
(±0.00073)
Rank: 65/77
0.42405
Rank: 63/77
Challenge organizers (contact) sift-def geodesc (128 float32: 512 bytes) GeoDesc descriptors extracted on SIFT keypoints and DEGENSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-23 is_baseline
Submission ID: 00043
DELF-GLD (128D), PyRANSAC
Size: 512 bytes. Matches: built-in
2036.82 94.70
Rank: 65/77
0.109
Rank: 73/77
0.156
Rank: 73/77
0.07298
(±0.00056)
Rank: 75/77
430.68
Rank: 7/77
2162.72
Rank: 5/77
2.535
Rank: 73/77
0.887
Rank: 74/77
0.13247
(±0.00116)
Rank: 72/77
0.10272
Rank: 73/77
Challenge organizers (contact) delf-gld-2k-128d delf-gld-2k-128d (128 float32: 512 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 128 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with PyRANSAC (DEGENSAC with the degeneracy check turned off). https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-23 is_baseline
Submission ID: ?????
CV-DoG-HardNet8-PTv2
Size: 512 bytes. Matches: built-in
2047.77 153.69
Rank: 30/77
0.339
Rank: 31/77
0.821
Rank: 6/77
0.43869
(±0.00017)
Rank: 22/77
158.63
Rank: 58/77
1221.12
Rank: 57/77
4.350
Rank: 27/77
0.484
Rank: 39/77
0.60206
(±0.00003)
Rank: 43/77
0.52038
Rank: 35/77
Milan Pultar, Dmytro Mishkin, Jiri Matas (contact) sift2k h8e512pt (128 float32: 512 bytes) [sid:00619] HardNet8 with PCA compression, batch sampling from few images N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00616
Sift-Fusion_Avg-NM-Net_End-to-En...
Size: 512 bytes. Matches: custom
1892.70 142.38
Rank: 37/77
0.333
Rank: 53/77
0.801
Rank: 24/77
0.44818
(±0.00000)
Rank: 19/77
146.58
Rank: 66/77
1339.45
Rank: 42/77
4.406
Rank: 25/77
0.456
Rank: 20/77
0.65563
(±0.00209)
Rank: 21/77
0.55190
Rank: 18/77
Chen Zhao (contact) siftdef fusion-avg (128 float32: 512 bytes) SIFT and Fusion_Avg, followed by NM-Net_End-to-End and DEGENSAC. N/A N/A 20-05-31 is_submission, is_challenge_2020
Submission ID: 00704
SOSNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 171.22
Rank: 21/77
0.333
Rank: 38/77
0.804
Rank: 22/77
0.45053
(±0.00053)
Rank: 18/77
194.04
Rank: 39/77
1442.32
Rank: 30/77
4.313
Rank: 36/77
0.467
Rank: 30/77
0.63586
(±0.00246)
Rank: 29/77
0.54320
Rank: 22/77
Challenge organizers (contact) sift-def sosnet-upright (128 float32: 512 bytes) SOSNet descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-06-02 is_baseline
Submission ID: 00710
DISK-32D (epi)
Size: 128 bytes. Matches: built-in
2048.00 352.43
Rank: 5/77
0.459
Rank: 3/77
0.826
Rank: 3/77
0.45946
(±0.00023)
Rank: 13/77
484.30
Rank: 2/77
2324.68
Rank: 2/77
5.589
Rank: 1/77
0.427
Rank: 9/77
0.69938
(±0.00192)
Rank: 9/77
0.57942
Rank: 7/77
Under review! NeurIPS anonymous submission, ID 1194 (To be released: 21-09-16) disk-2020-09-15-nms-5-epi-32-save-49-imsize-1024-nump-2048 disk-2020-09-15-nms-5-epi-32-save-49-imsize-1024-nump-2048 (32 float32: 128 bytes) Local feature model learned via policy gradient, using 32D descriptors. Model trained with a cycle-consistency loss and supervised with epipolar constraints. Trained on MegaDepth, removing conflicts with the test data. For inference, images are resized to 1024 pixels on the longest edge, with NMS over a 5x5 window, taking the top 2048 features by score. N/A N/A 20-09-16 is_baseline, is_under_review
Submission ID: 00578
Key.Net-s + DescNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
2033.67 246.62
Rank: 12/77
0.449
Rank: 5/77
0.805
Rank: 21/77
0.45418
(±0.00056)
Rank: 15/77
331.55
Rank: 13/77
1621.68
Rank: 21/77
4.570
Rank: 12/77
0.447
Rank: 15/77
0.67414
(±0.00371)
Rank: 15/77
0.56416
Rank: 14/77
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) keynet-s descnet (128 float32: 512 bytes) N/A N/A 20-05-20 is_submission, is_challenge_2020
Submission ID: 00030
Upright Root-SIFT (OpenCV), DEGE...
Size: 512 bytes. Matches: built-in
1892.72 111.69
Rank: 56/77
0.333
Rank: 34/77
0.781
Rank: 39/77
0.39547
(±0.00027)
Rank: 43/77
201.15
Rank: 36/77
1347.75
Rank: 40/77
4.091
Rank: 52/77
0.522
Rank: 51/77
0.56078
(±0.00264)
Rank: 53/77
0.47812
Rank: 48/77
Challenge organizers (contact) sift-def rootsift-upright (128 float32: 512 bytes) Upright Root-SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00009
GeoDesc (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 116.56
Rank: 50/77
0.333
Rank: 38/77
0.784
Rank: 34/77
0.38660
(±0.00078)
Rank: 47/77
149.66
Rank: 64/77
1237.96
Rank: 54/77
4.219
Rank: 48/77
0.525
Rank: 53/77
0.56687
(±0.00261)
Rank: 50/77
0.47673
Rank: 49/77
Challenge organizers (contact) sift-def geodesc-upright (128 float32: 512 bytes) GeoDesc descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00701
GeoDesc (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 132.69
Rank: 44/77
0.333
Rank: 38/77
0.798
Rank: 27/77
0.41364
(±0.00033)
Rank: 35/77
161.05
Rank: 57/77
1287.26
Rank: 50/77
4.239
Rank: 43/77
0.502
Rank: 46/77
0.58373
(±0.00402)
Rank: 47/77
0.49868
Rank: 41/77
Challenge organizers (contact) sift-def geodesc-upright (128 float32: 512 bytes) GeoDesc descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-06-01 is_baseline
Submission ID: 00017
ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
2031.83 63.43
Rank: 74/77
0.366
Rank: 20/77
0.523
Rank: 72/77
0.08243
(±0.00058)
Rank: 73/77
120.93
Rank: 73/77
280.38
Rank: 77/77
2.191
Rank: 77/77
0.863
Rank: 72/77
0.02318
(±0.00229)
Rank: 77/77
0.05280
Rank: 77/77
Challenge organizers (contact) orb orb (32 uint8: 32 bytes) ORB with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00589
SuperPoint-128d + SuperGlue + DE...
Size: 512 bytes. Matches: custom
1973.62 320.46
Rank: 7/77
0.364
Rank: 21/77
0.772
Rank: 47/77
0.55214
(±0.00098)
Rank: 3/77
429.49
Rank: 9/77
2130.73
Rank: 8/77
4.570
Rank: 11/77
0.384
Rank: 3/77
0.75283
(±0.00086)
Rank: 3/77
0.65248
Rank: 3/77
Paul-Edouard Sarlin (contact) superpoint-k2048-nms3-refine2-r1600forcecubic superpoint-down128 (128 float32: 512 bytes) SuperPoint detector (2048 keypoints, NMS with radius 3, refinement, on 1600-pixel images) and descriptor; reduced to 128d with a linear autoencoder. SuperGlue matcher (outdoor model, 150 Sinkhorn iterations). For stereo, DEGENSAC model estimator (1.2 pixels inlier threshold). https://arxiv.org/abs/1911.11763 https://psarlin.com/superglue 20-05-24 is_submission, is_challenge_2020
Submission ID: 00702
L2-Net (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 117.11
Rank: 47/77
0.333
Rank: 38/77
0.808
Rank: 17/77
0.41918
(±0.00059)
Rank: 34/77
179.79
Rank: 45/77
1361.34
Rank: 39/77
4.232
Rank: 44/77
0.481
Rank: 38/77
0.59682
(±0.00079)
Rank: 44/77
0.50800
Rank: 39/77
Challenge organizers (contact) sift-def l2net-upright (128 float32: 512 bytes) L2-Net descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. http://www.nlpr.ia.ac.cn/fanbin/pub/L2-Net_CVPR17.pdf https://github.com/vcg-uvic/image-matching-benchmark-baselines 20-06-01 is_baseline
Submission ID: 00603
SuperPoint-128d-masked + SuperGl...
Size: 512 bytes. Matches: custom
2041.09 404.73
Rank: 2/77
0.387
Rank: 16/77
0.774
Rank: 45/77
0.56769
(±0.00034)
Rank: 2/77
415.75
Rank: 10/77
2158.03
Rank: 7/77
4.932
Rank: 5/77
0.357
Rank: 1/77
0.76987
(±0.00122)
Rank: 2/77
0.66878
Rank: 2/77
Paul-Edouard Sarlin (contact) superpoint-k2048-nms3-refine2-r1600forcecubic-masked-d.001 superpoint-down128 (128 float32: 512 bytes) SuperPoint detector (2048 keypoints, NMS with radius 3, confidence threshold 0.001, refinement, on 1600-pixel images). Detections on semantic classes sky and people are removed (segmentation from HFNetV2 trained on MIT ADE20K). SuperPoint descriptor, reduced to 128d with a linear autoencoder. SuperGlue matcher (outdoor model, 150 Sinkhorn iterations). For stereo, DEGENSAC model estimator (1.2 pixel inlier threshold). https://arxiv.org/abs/1911.11763 https://psarlin.com/superglue 20-05-30 is_submission, is_challenge_2020
Submission ID: 00609
Key.Net-s-ref + HyNet w/ DEGENSA...
Size: 512 bytes. Matches: built-in
2032.93 270.04
Rank: 9/77
0.461
Rank: 2/77
0.820
Rank: 8/77
0.46902
(±0.00032)
Rank: 7/77
280.40
Rank: 19/77
1489.62
Rank: 25/77
4.689
Rank: 7/77
0.436
Rank: 12/77
0.68120
(±0.00119)
Rank: 12/77
0.57511
Rank: 11/77
Barroso-Laguna, Axel and Tian, Yurun, Ng, Tony and Mikolajczyk, Krystian (contact) keynet-s-ref hynet (128 float32: 512 bytes) N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00036
SOSNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 155.75
Rank: 28/77
0.333
Rank: 38/77
0.794
Rank: 30/77
0.43077
(±0.00017)
Rank: 27/77
183.92
Rank: 42/77
1403.66
Rank: 34/77
4.314
Rank: 34/77
0.472
Rank: 33/77
0.62181
(±0.00161)
Rank: 36/77
0.52629
Rank: 34/77
Challenge organizers (contact) sift-def sosnet-upright (128 float32: 512 bytes) SOSNet descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00615
SIFT-Fusion_Max-NM-Net_End-to-En...
Size: 512 bytes. Matches: custom
1892.70 115.46
Rank: 52/77
0.333
Rank: 53/77
0.783
Rank: 35/77
0.40244
(±0.00000)
Rank: 39/77
119.23
Rank: 76/77
1167.54
Rank: 63/77
4.367
Rank: 26/77
0.489
Rank: 40/77
0.60720
(±0.00114)
Rank: 39/77
0.50482
Rank: 40/77
Chen Zhao (contact) siftdef fusion-max (128 float32: 512 bytes) SIFT and Fusion_Max, followed by NM-Net_End-to-End and DEGENSAC. N/A N/A 20-05-31 is_submission, is_challenge_2020
Submission ID: 00012
GeoDesc (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 144.92
Rank: 36/77
0.333
Rank: 38/77
0.765
Rank: 52/77
0.38124
(±0.00056)
Rank: 53/77
149.66
Rank: 64/77
1237.96
Rank: 54/77
4.219
Rank: 48/77
0.526
Rank: 55/77
0.56687
(±0.00261)
Rank: 50/77
0.47405
Rank: 50/77
Challenge organizers (contact) sift-def geodesc-upright (128 float32: 512 bytes) GeoDesc descriptors extracted on SIFT keypoints with a fixed orientation, and MAGSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-23 is_baseline
Submission ID: 00026
AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
1657.83 69.77
Rank: 73/77
0.355
Rank: 23/77
0.622
Rank: 70/77
0.17632
(±0.00087)
Rank: 71/77
224.40
Rank: 30/77
795.54
Rank: 76/77
3.311
Rank: 71/77
0.819
Rank: 71/77
0.23020
(±0.00466)
Rank: 71/77
0.20326
Rank: 71/77
Challenge organizers (contact) akaze-def akaze (61 uint8: 61 bytes) AKAZE with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00583
R2D2
Size: 512 bytes. Matches: built-in
2048.00 169.38
Rank: 22/77
0.398
Rank: 15/77
0.695
Rank: 65/77
0.33715
(±0.00149)
Rank: 62/77
306.15
Rank: 15/77
1235.83
Rank: 56/77
4.156
Rank: 51/77
0.496
Rank: 42/77
0.60223
(±0.00103)
Rank: 42/77
0.46969
Rank: 55/77
(contact) r2d2-5k-p-aug r2d2-5k-p-aug (128 float32: 512 bytes) N/A N/A 20-05-23 is_submission, is_challenge_2020
Submission ID: 00513
Personal Inplementation SIFT (GP...
Size: 512 bytes. Matches: built-in
1900.60 70.04
Rank: 72/77
0.085
Rank: 77/77
0.002
Rank: 77/77
0.07782
(±0.00022)
Rank: 74/77
173.06
Rank: 47/77
948.78
Rank: 73/77
3.816
Rank: 67/77
0.887
Rank: 75/77
0.09835
(±0.00068)
Rank: 74/77
0.08809
Rank: 74/77
feyman_priv (contact) sift-gpu sift-gpu (128 float32: 512 bytes) SIFT with 2048 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00013
SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1936.30 76.94
Rank: 70/77
0.323
Rank: 68/77
0.733
Rank: 62/77
0.30766
(±0.00075)
Rank: 66/77
171.92
Rank: 50/77
1059.70
Rank: 72/77
3.687
Rank: 69/77
0.618
Rank: 68/77
0.42476
(±0.00103)
Rank: 68/77
0.36621
Rank: 68/77
Challenge organizers (contact) sift-def sift (128 float32: 512 bytes) SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-23 is_baseline
Submission ID: 00706
LogPolar-Upright w/ DEGENSAC (no...
Size: 512 bytes. Matches: built-in
1892.71 162.20
Rank: 23/77
0.333
Rank: 38/77
0.807
Rank: 19/77
0.45674
(±0.00059)
Rank: 14/77
211.87
Rank: 31/77
1553.36
Rank: 23/77
4.331
Rank: 31/77
0.471
Rank: 32/77
0.63701
(±0.00278)
Rank: 28/77
0.54688
Rank: 20/77
Challenge organizers (contact) sift-def logpolar-upright (128 float32: 512 bytes) Upright LogPolar descriptors on DoG features (OpenCV). Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-06-02 is_baseline
Submission ID: 00123
KeyNet-HardNet-2k
Size: 512 bytes. Matches: built-in
2048.00 134.39
Rank: 42/77
0.469
Rank: 1/77
0.818
Rank: 9/77
0.32725
(±0.00026)
Rank: 64/77
195.33
Rank: 38/77
1276.30
Rank: 51/77
4.493
Rank: 19/77
0.490
Rank: 41/77
0.61606
(±0.00141)
Rank: 38/77
0.47166
Rank: 52/77
Challenge organizers (contact) keynettuned vlhardnet (128 float32: 512 bytes) KeyNet-HardNet with 2048 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-03 is_baseline
Submission ID: 00608
SIFT2k_2048_HardNet64-train-all-...
Size: 512 bytes. Matches: custom
2047.00 245.40
Rank: 13/77
0.346
Rank: 27/77
0.823
Rank: 4/77
0.49219
(±0.00000)
Rank: 5/77
253.37
Rank: 25/77
1984.40
Rank: 9/77
4.607
Rank: 9/77
0.421
Rank: 8/77
0.70020
(±0.00213)
Rank: 8/77
0.59619
Rank: 5/77
Ximin Zheng, Sheng He, Hualong Shi (contact) sift2k hardnet64-train-all-l2-138000-matched (128 float32: 512 bytes) SIFT with 2048 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 138000), FLANN disabled, custom matches N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00044
Upright SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.72 104.22
Rank: 62/77
0.333
Rank: 34/77
0.763
Rank: 54/77
0.37078
(±0.00056)
Rank: 56/77
205.17
Rank: 33/77
1300.49
Rank: 48/77
4.000
Rank: 64/77
0.563
Rank: 65/77
0.52549
(±0.00120)
Rank: 62/77
0.44814
Rank: 60/77
Challenge organizers (contact) sift-def sift-upright (128 float32: 512 bytes) Upright SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00503
Guided matching hardnet
Size: 512 bytes. Matches: custom
1892.71 149.75
Rank: 34/77
0.333
Rank: 52/77
0.795
Rank: 29/77
0.42159
(±0.00000)
Rank: 33/77
254.13
Rank: 23/77
1754.99
Rank: 16/77
4.283
Rank: 42/77
0.464
Rank: 26/77
0.64144
(±0.00207)
Rank: 27/77
0.53152
Rank: 30/77
(contact) sift upright-hardnet (128 float32: 512 bytes) Sift keypoint ; upright hardnet descriptors ; custom matching: use a deep learning based coarse matcher as a first step. In a second step match the keypoint according to descriptor distances but only for the matches that are close to the coarse match prediction N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00022
L2-Net (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 110.72
Rank: 57/77
0.333
Rank: 38/77
0.801
Rank: 25/77
0.40729
(±0.00097)
Rank: 36/77
171.93
Rank: 48/77
1333.56
Rank: 43/77
4.226
Rank: 46/77
0.497
Rank: 44/77
0.58762
(±0.00228)
Rank: 45/77
0.49746
Rank: 42/77
Challenge organizers (contact) sift-def l2net-upright (128 float32: 512 bytes) L2-Net descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. http://www.nlpr.ia.ac.cn/fanbin/pub/L2-Net_CVPR17.pdf https://github.com/vcg-uvic/image-matching-benchmark-baselines 20-04-23 is_baseline
Submission ID: 00016
LogPolarDesc, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 148.56
Rank: 35/77
0.323
Rank: 58/77
0.752
Rank: 60/77
0.38840
(±0.00064)
Rank: 46/77
162.84
Rank: 55/77
1385.60
Rank: 37/77
4.046
Rank: 55/77
0.516
Rank: 48/77
0.57149
(±0.00344)
Rank: 48/77
0.47994
Rank: 47/77
Challenge organizers (contact) sift-def logpolar (128 float32: 512 bytes) LogPolarDesc descriptors extracted on SIFT keypoints and MAGSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-23 is_baseline
Submission ID: 00622
Sift-HardNet-NM-Net_End-to-End
Size: 512 bytes. Matches: custom
1892.70 152.13
Rank: 32/77
0.333
Rank: 53/77
0.807
Rank: 20/77
0.46620
(±0.00000)
Rank: 9/77
156.54
Rank: 59/77
1403.97
Rank: 33/77
4.425
Rank: 23/77
0.443
Rank: 14/77
0.67160
(±0.00184)
Rank: 16/77
0.56890
Rank: 12/77
Chen Zhao (contact) siftdef hardnet (128 float32: 512 bytes) SIFT and HardNet, followed by NM-Net_End-to-End and DEGENSAC. N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00032
Root-SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1936.30 83.46
Rank: 69/77
0.323
Rank: 68/77
0.756
Rank: 59/77
0.33605
(±0.00092)
Rank: 63/77
167.84
Rank: 52/77
1141.88
Rank: 66/77
3.788
Rank: 68/77
0.576
Rank: 67/77
0.46826
(±0.00154)
Rank: 67/77
0.40216
Rank: 66/77
Challenge organizers (contact) sift-def rootsift (128 float32: 512 bytes) Root-SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00040
DELF-GLD (128D), DEGENSAC
Size: 512 bytes. Matches: built-in
2036.82 100.78
Rank: 63/77
0.109
Rank: 73/77
0.151
Rank: 74/77
0.08367
(±0.00047)
Rank: 72/77
430.68
Rank: 7/77
2162.72
Rank: 5/77
2.535
Rank: 73/77
0.880
Rank: 73/77
0.13247
(±0.00116)
Rank: 72/77
0.10807
Rank: 72/77
Challenge organizers (contact) delf-gld-2k-128d delf-gld-2k-128d (128 float32: 512 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 128 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with DEGENSAC. https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-22 is_baseline
Submission ID: 00605
guided-hardnet-epoch2-v2
Size: 512 bytes. Matches: custom
2047.76 220.88
Rank: 15/77
0.347
Rank: 24/77
0.778
Rank: 41/77
0.46469
(±0.00000)
Rank: 10/77
226.84
Rank: 29/77
1827.60
Rank: 13/77
4.512
Rank: 16/77
0.427
Rank: 10/77
0.69306
(±0.00160)
Rank: 11/77
0.57888
Rank: 10/77
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift2k hardnet-epoch2 (128 float32: 512 bytes) sift and hardnet with 2k features, using the oanet and guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-30 is_submission, is_challenge_2020
Submission ID: 00027
DELF-GLD (32D), PyRANSAC
Size: 128 bytes. Matches: built-in
2036.82 91.18
Rank: 67/77
0.109
Rank: 73/77
0.122
Rank: 75/77
0.06197
(±0.00032)
Rank: 77/77
453.19
Rank: 3/77
1805.48
Rank: 14/77
2.425
Rank: 75/77
0.916
Rank: 76/77
0.09163
(±0.00140)
Rank: 75/77
0.07680
Rank: 76/77
Challenge organizers (contact) delf-gld-2k-32d delf-gld-2k-32d (32 float32: 128 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 32 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with PyRANSAC (DEGENSAC with the degeneracy check turned off). https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-22 is_baseline
Submission ID: 00037
GeoDesc, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 115.17
Rank: 53/77
0.323
Rank: 64/77
0.742
Rank: 61/77
0.34024
(±0.00050)
Rank: 61/77
120.68
Rank: 74/77
1098.70
Rank: 68/77
4.002
Rank: 62/77
0.551
Rank: 63/77
0.50594
(±0.00073)
Rank: 65/77
0.42309
Rank: 64/77
Challenge organizers (contact) sift-def geodesc (128 float32: 512 bytes) GeoDesc descriptors extracted on SIFT keypoints and MAGSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-23 is_baseline
Submission ID: 00023
LogPolarDesc, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 116.83
Rank: 48/77
0.323
Rank: 58/77
0.770
Rank: 48/77
0.39335
(±0.00060)
Rank: 44/77
162.84
Rank: 55/77
1385.60
Rank: 37/77
4.046
Rank: 55/77
0.519
Rank: 50/77
0.57149
(±0.00344)
Rank: 48/77
0.48242
Rank: 44/77
Challenge organizers (contact) sift-def logpolar (128 float32: 512 bytes) LogPolarDesc descriptors extracted on SIFT keypoints and DEGENSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00564
SIFT2k_2000_HardNet64_train_all_...
Size: 512 bytes. Matches: built-in
1999.83 156.93
Rank: 27/77
0.336
Rank: 32/77
0.814
Rank: 11/77
0.42959
(±0.00060)
Rank: 28/77
244.13
Rank: 28/77
1646.76
Rank: 20/77
4.334
Rank: 30/77
0.459
Rank: 23/77
0.64407
(±0.00339)
Rank: 24/77
0.53683
Rank: 27/77
Ximin Zheng, Sheng He, Hualong Shi (contact) sift2k sift2k-2000-hardnet64-train-all-l2-138000 (128 float32: 512 bytes) SIFT with 2000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 138000), FLANN disabled N/A N/A 20-05-13 is_submission, is_challenge_2020
Submission ID: 00025
DELF-GLD (32D), DEGENSAC
Size: 128 bytes. Matches: built-in
2036.82 96.56
Rank: 64/77
0.109
Rank: 73/77
0.119
Rank: 76/77
0.06573
(±0.00030)
Rank: 76/77
453.19
Rank: 3/77
1805.48
Rank: 14/77
2.425
Rank: 75/77
0.920
Rank: 77/77
0.09163
(±0.00140)
Rank: 75/77
0.07868
Rank: 75/77
Challenge organizers (contact) delf-gld-2k-32d delf-gld-2k-32d (32 float32: 128 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 32 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with DEGENSAC. https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-22 is_baseline
Submission ID: 00512
Hardnet with DEGENSAC for stereo
Size: 512 bytes. Matches: built-in
1999.50 51.91
Rank: 77/77
0.172
Rank: 72/77
0.558
Rank: 71/77
0.20771
(±0.00049)
Rank: 69/77
181.48
Rank: 44/77
1692.63
Rank: 18/77
3.050
Rank: 72/77
0.695
Rank: 69/77
0.38463
(±0.00275)
Rank: 69/77
0.29617
Rank: 69/77
Vu Trung Nghia & Nguyen Trung Hieu (contact) siftdef hardnet-64 (128 float32: 512 bytes) Using hardnet network to embed a patch (32 x 32) to a 128 (float32) dimensions vector, For stereo, we use DEGENSAC (Chum et al, CVPR'05) with optimal settings N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00527
SEKD
Size: 512 bytes. Matches: built-in
1786.69 106.34
Rank: 60/77
0.383
Rank: 19/77
0.759
Rank: 57/77
0.38886
(±0.00013)
Rank: 45/77
150.11
Rank: 63/77
1063.08
Rank: 71/77
4.575
Rank: 10/77
0.465
Rank: 28/77
0.63303
(±0.00244)
Rank: 31/77
0.51095
Rank: 38/77
Yafei Song, Ling Cai, Mingyang Li (contact) sekd sekd (128 float32: 512 bytes) The name of our method is Self-Evolving Keypoint Detection and Description (SEKD). Now, the SEKD model is trained only using COCO test images. In this submission each image has up to 2048 SEKD keypoints, and 128-dim float descriptor. We use the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC. N/A N/A 20-04-27 is_submission, is_challenge_2020
Submission ID: 00711
DISK-32D (depth)
Size: 128 bytes. Matches: built-in
2048.00 341.83
Rank: 6/77
0.448
Rank: 6/77
0.842
Rank: 2/77
0.47005
(±0.00037)
Rank: 6/77
441.16
Rank: 6/77
2231.55
Rank: 4/77
5.524
Rank: 3/77
0.415
Rank: 5/77
0.70220
(±0.00208)
Rank: 7/77
0.58613
Rank: 6/77
Under review! NeurIPS anonymous submission, ID 1194 (To be released: 21-09-16) disk-2020-09-15-nms-7-depth-32-save-46-imsize-1024-nump-2048 disk-2020-09-15-nms-7-depth-32-save-46-imsize-1024-nump-2048 (32 float32: 128 bytes) Local feature model learned via policy gradient, using 32D descriptors. Model trained with a cycle-consistency loss and supervised with depth-based constraints. Trained on MegaDepth, removing conflicts with the test data. For inference, images are resized to 1024 pixels on the longest edge, with NMS over a 7x7 window, taking the top 2048 features by score. N/A N/A 20-09-16 is_baseline, is_under_review
Submission ID: 00612
SuperPoint-128d-adapt + SuperGlu...
Size: 512 bytes. Matches: custom
2048.00 441.49
Rank: 1/77
0.407
Rank: 13/77
0.789
Rank: 32/77
0.59034
(±0.00050)
Rank: 1/77
452.99
Rank: 5/77
2245.43
Rank: 3/77
5.092
Rank: 4/77
0.358
Rank: 2/77
0.77337
(±0.00167)
Rank: 1/77
0.68186
Rank: 1/77
Paul-Edouard Sarlin (contact) superpoint-k2048-nms4-refine2-r1600forcecubic-masked-d.001-adapt50 superpoint-down128 (128 float32: 512 bytes) SuperPoint detector (2048 keypoints, NMS with radius 4, confidence threshold 0.001, refinement, on 1600-pixel images). The detection heatmap is improved with test-time homographic adaptation (50 iterations), and detections on semantic classes sky and people are removed (segmentation from HFNetV2 trained on MIT ADE20K). SuperPoint descriptor, reduced to 128d with a linear autoencoder. SuperGlue matcher (outdoor model, 150 Sinkhorn iterations). For stereo, DEGENSAC model estimator (1.1 pixel inlier threshold). https://arxiv.org/abs/1911.11763 https://psarlin.com/superglue 20-05-31 is_submission, is_challenge_2020
Submission ID: 00586
KeyNet+HardNet+NM-Net_v2
Size: 512 bytes. Matches: custom
2042.89 160.25
Rank: 25/77
0.435
Rank: 11/77
0.821
Rank: 5/77
0.40264
(±0.00000)
Rank: 38/77
165.01
Rank: 54/77
1134.69
Rank: 67/77
4.720
Rank: 6/77
0.448
Rank: 17/77
0.67488
(±0.00197)
Rank: 14/77
0.53876
Rank: 24/77
Chen Zhao (contact) keynet hardnet (128 float32: 512 bytes) KeyNet and HardNet, followed by NM-Net_v2 and DEGENSAC. N/A N/A 20-05-23 is_submission, is_challenge_2020
Submission ID: 00585
NM-Net_v2
Size: 512 bytes. Matches: custom
1892.70 151.46
Rank: 33/77
0.333
Rank: 53/77
0.809
Rank: 15/77
0.46620
(±0.00000)
Rank: 8/77
155.91
Rank: 60/77
1392.47
Rank: 36/77
4.424
Rank: 24/77
0.459
Rank: 22/77
0.66837
(±0.00125)
Rank: 17/77
0.56729
Rank: 13/77
Chen Zhao (contact) siftdef hardnet (128 float32: 512 bytes) SIFT and HardNet, followed by NM-Net_v2 and DEGENSAC. N/A N/A 20-05-23 is_submission, is_challenge_2020
Submission ID: 00517
retrained Hardnet with MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 133.39
Rank: 43/77
0.333
Rank: 38/77
0.808
Rank: 18/77
0.40042
(±0.00077)
Rank: 40/77
210.10
Rank: 32/77
1480.67
Rank: 26/77
4.226
Rank: 45/77
0.479
Rank: 37/77
0.63030
(±0.00242)
Rank: 32/77
0.51536
Rank: 36/77
Vu Trung Nghia and Nguyen Trung Hieu (contact) sift-upright hardnet (128 float32: 512 bytes) Using SIFT upright and retrained Hardnet for keypoints and descriptors. For stereo, we use DEGENSAC Chum et al, CVPR05 with optimal settings N/A N/A 20-04-24 is_submission, is_challenge_2020
Submission ID: 00509
Upright-Sift + X-Net-lib w/ DEGE...
Size: 512 bytes. Matches: built-in
1892.70 116.55
Rank: 51/77
0.333
Rank: 57/77
0.821
Rank: 7/77
0.42382
(±0.00089)
Rank: 30/77
168.67
Rank: 51/77
1319.63
Rank: 45/77
4.303
Rank: 38/77
0.468
Rank: 31/77
0.62953
(±0.00216)
Rank: 33/77
0.52668
Rank: 33/77
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) sift-def x-net-lib-upright-no-dups (128 float32: 512 bytes) N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: ?????
CV-DoG-HardNet8-PT
Size: 512 bytes. Matches: built-in
2048.00 114.84
Rank: 54/77
0.321
Rank: 71/77
0.790
Rank: 31/77
0.38200
(±0.00081)
Rank: 52/77
117.84
Rank: 77/77
1067.13
Rank: 70/77
4.051
Rank: 54/77
0.531
Rank: 56/77
0.53035
(±0.00531)
Rank: 60/77
0.45617
Rank: 58/77
Milan Pultar, Dmytro Mishkin, Jiri Matas (contact) sift2k h8e512pt (128 float32: 512 bytes) [sid:00593] HardNet8 with PCA compression N/A N/A 20-05-27 is_submission, is_challenge_2020
Submission ID: 00003
HardNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 141.66
Rank: 38/77
0.333
Rank: 38/77
0.803
Rank: 23/77
0.44235
(±0.00127)
Rank: 21/77
193.09
Rank: 40/77
1435.93
Rank: 31/77
4.302
Rank: 39/77
0.474
Rank: 35/77
0.62275
(±0.00423)
Rank: 34/77
0.53255
Rank: 29/77
Challenge organizers (contact) sift-def hardnet-upright (128 float32: 512 bytes) HardNet descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00035
HardNet (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 181.81
Rank: 17/77
0.333
Rank: 38/77
0.781
Rank: 38/77
0.43816
(±0.00058)
Rank: 23/77
193.09
Rank: 40/77
1435.93
Rank: 31/77
4.302
Rank: 39/77
0.473
Rank: 34/77
0.62275
(±0.00423)
Rank: 34/77
0.53045
Rank: 31/77
Challenge organizers (contact) sift-def hardnet-upright (128 float32: 512 bytes) HardNet descriptors extracted on SIFT keypoints with a fixed orientation, and MAGSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00708
DISK (LCC/depth)
Size: 512 bytes. Matches: built-in
2048.00 404.19
Rank: 3/77
0.448
Rank: 7/77
0.852
Rank: 1/77
0.51315
(±0.00028)
Rank: 4/77
527.48
Rank: 1/77
2428.04
Rank: 1/77
5.545
Rank: 2/77
0.410
Rank: 4/77
0.72705
(±0.00094)
Rank: 4/77
0.62010
Rank: 4/77
Under review! NeurIPS anonymous submission, ID 1194 (To be released: 21-06-03) disk-cc-continued-20-imsize-1024-nms-3-nump-2048 disk-cc-continued-20-imsize-1024-nms-3-nump-2048 (128 float32: 512 bytes) Local feature model learned via policy gradient. Model trained with a cycle-consistency loss and supervised with depth. Trained on MegaDepth, removing conflicts with the test data. For inference, images are resized to 1024 pixels on the longest edge, with NMS over a 3x3 window. We take the top 2048 features by score. N/A N/A 20-06-03 is_submission, is_challenge_2020, is_under_review
Submission ID: 00004
SOSNet (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 176.54
Rank: 18/77
0.333
Rank: 38/77
0.782
Rank: 36/77
0.43803
(±0.00092)
Rank: 24/77
183.92
Rank: 42/77
1403.66
Rank: 34/77
4.314
Rank: 34/77
0.474
Rank: 36/77
0.62181
(±0.00161)
Rank: 36/77
0.52992
Rank: 32/77
Challenge organizers (contact) sift-def sosnet-upright (128 float32: 512 bytes) SOSNet descriptors extracted on SIFT keypoints with a fixed orientation, and MAGSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00018
R2D2 (r2d2-wasf-n16), DEGENSAC
Size: 512 bytes. Matches: built-in
2048.00 176.49
Rank: 19/77
0.422
Rank: 12/77
0.725
Rank: 63/77
0.35783
(±0.00061)
Rank: 58/77
262.02
Rank: 21/77
1188.11
Rank: 62/77
4.299
Rank: 41/77
0.499
Rank: 45/77
0.60639
(±0.00124)
Rank: 40/77
0.48211
Rank: 45/77
Challenge organizers (contact) r2d2-wasf-n16 r2d2-wasf-n16 (128 float32: 512 bytes) R2D2 (model r2d2-wasf-n16) with 2k features, using the built-in matcher (bidirectional filter with the both strategy) and no ratio test, and DEGENSAC with optimal parameters for stereo. N/A https://opencv.org 20-04-23 is_baseline
Submission ID: 00014
L2-Net, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 107.38
Rank: 59/77
0.323
Rank: 64/77
0.766
Rank: 50/77
0.35649
(±0.00110)
Rank: 59/77
138.97
Rank: 69/77
1211.49
Rank: 58/77
3.948
Rank: 65/77
0.538
Rank: 61/77
0.52019
(±0.00120)
Rank: 63/77
0.43834
Rank: 62/77
Challenge organizers (contact) sift-def l2net (128 float32: 512 bytes) L2-Net descriptors extracted on SIFT keypoints and MAGSAC for stereo. Please refer to the baselines repository (linked) for details. http://www.nlpr.ia.ac.cn/fanbin/pub/L2-Net_CVPR17.pdf https://github.com/vcg-uvic/image-matching-benchmark-baselines 20-04-23 is_baseline
Submission ID: 00700
HardNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 152.69
Rank: 31/77
0.333
Rank: 38/77
0.812
Rank: 14/77
0.46093
(±0.00080)
Rank: 12/77
201.26
Rank: 35/77
1467.86
Rank: 29/77
4.311
Rank: 37/77
0.466
Rank: 29/77
0.63544
(±0.00324)
Rank: 30/77
0.54819
Rank: 19/77
Challenge organizers (contact) sift-def hardnet-upright (128 float32: 512 bytes) HardNet descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-06-01 is_baseline
Submission ID: 00033
SOSNet, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 119.91
Rank: 46/77
0.323
Rank: 58/77
0.769
Rank: 49/77
0.37632
(±0.00056)
Rank: 54/77
145.47
Rank: 67/77
1271.41
Rank: 52/77
4.045
Rank: 57/77
0.524
Rank: 52/77
0.56072
(±0.00077)
Rank: 54/77
0.46852
Rank: 56/77
Challenge organizers (contact) sift-def sosnet (128 float32: 512 bytes) SOSNet descriptors extracted on SIFT keypoints and DEGENSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00039
HardNet, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 138.43
Rank: 40/77
0.323
Rank: 58/77
0.757
Rank: 58/77
0.38490
(±0.00120)
Rank: 51/77
153.29
Rank: 61/77
1306.64
Rank: 46/77
4.026
Rank: 59/77
0.537
Rank: 60/77
0.55573
(±0.00175)
Rank: 58/77
0.47032
Rank: 54/77
Challenge organizers (contact) sift-def hardnet (128 float32: 512 bytes) HardNet descriptors extracted on SIFT keypoints and MAGSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00582
R2D2
Size: 512 bytes. Matches: built-in
2048.00 155.21
Rank: 29/77
0.403
Rank: 14/77
0.709
Rank: 64/77
0.32501
(±0.00085)
Rank: 65/77
285.11
Rank: 18/77
1162.94
Rank: 65/77
4.161
Rank: 50/77
0.508
Rank: 47/77
0.60276
(±0.00062)
Rank: 41/77
0.46389
Rank: 57/77
(contact) r2d2-5k-p r2d2-5k-p (128 float32: 512 bytes) N/A N/A 20-05-23 is_submission, is_challenge_2020
Submission ID: 00617
pffNet + SuperPoint + DEGENSAC
Size: 512 bytes. Matches: built-in
1267.22 55.58
Rank: 76/77
0.341
Rank: 28/77
0.629
Rank: 69/77
0.25514
(±0.00059)
Rank: 67/77
131.91
Rank: 71/77
893.88
Rank: 74/77
4.339
Rank: 28/77
0.534
Rank: 58/77
0.55935
(±0.00143)
Rank: 56/77
0.40724
Rank: 65/77
Jongmin Lee, Seungwook Kim, Yoonwoo Jeong (contact) superpoint pffnet (128 float32: 512 bytes) pffNet descriptors+ SuperPoint keypoints + DEGENSAC outlier-filtering N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00008
L2-Net (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 138.44
Rank: 39/77
0.333
Rank: 38/77
0.786
Rank: 33/77
0.39980
(±0.00020)
Rank: 41/77
171.93
Rank: 48/77
1333.56
Rank: 43/77
4.226
Rank: 46/77
0.496
Rank: 43/77
0.58762
(±0.00228)
Rank: 45/77
0.49371
Rank: 43/77
Challenge organizers (contact) sift-def l2net-upright (128 float32: 512 bytes) L2-Net descriptors extracted on SIFT keypoints with a fixed orientation, and MAGSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. http://www.nlpr.ia.ac.cn/fanbin/pub/L2-Net_CVPR17.pdf https://github.com/vcg-uvic/image-matching-benchmark-baselines 20-04-22 is_baseline
Submission ID: 00531
Key.Net + X-Net w/ DEGENSAC
Size: 512 bytes. Matches: built-in
2040.08 173.65
Rank: 20/77
0.447
Rank: 8/77
0.797
Rank: 28/77
0.38608
(±0.00068)
Rank: 49/77
276.91
Rank: 20/77
1468.50
Rank: 28/77
4.447
Rank: 20/77
0.463
Rank: 25/77
0.64369
(±0.00081)
Rank: 26/77
0.51489
Rank: 37/77
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) keynet x-net (128 float32: 512 bytes) N/A N/A 20-04-24 is_submission, is_challenge_2020

Phototourism: unlimited keypoints, small descriptors (128 bytes)

Note: entries with the same multi-view configuration may seem duplicated. This is normal: performance is averaged across tasks.
Stereo Multiview Avg.
Method NF NI Rep.
(3 pix.)
MS
(3 pix.)
mAA
(at 10o)
NM NL TL ATE mAA
(at 100)
mAA
(at 100)
Authors Keypoint Descriptor Summary Paper Website Processing date Flags
Submission ID: 00042
ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
7150.21 161.98
Rank: 2/2
0.514
Rank: 2/2
0.653
Rank: 2/2
0.16159
(±0.00090)
Rank: 2/2
910.31
Rank: 1/2
1423.38
Rank: 2/2
2.722
Rank: 2/2
0.897
Rank: 2/2
0.08054
(±0.00242)
Rank: 2/2
0.12106
Rank: 2/2
Challenge organizers (contact) orb orb (32 uint8: 32 bytes) ORB with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00010
AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
7857.11 246.74
Rank: 1/2
0.553
Rank: 1/2
0.735
Rank: 1/2
0.30717
(±0.00122)
Rank: 1/2
479.55
Rank: 2/2
2778.68
Rank: 1/2
3.393
Rank: 1/2
0.737
Rank: 1/2
0.36048
(±0.00382)
Rank: 1/2
0.33383
Rank: 1/2
Challenge organizers (contact) akaze-lowth akaze (61 uint8: 61 bytes) AKAZE with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline

Phototourism: restricted keypoints, small descriptors (128 bytes)

Note: entries with the same multi-view configuration may seem duplicated. This is normal: performance is averaged across tasks.
Stereo Multiview Avg.
Method NF NI Rep.
(3 pix.)
MS
(3 pix.)
mAA
(at 10o)
NM NL TL ATE mAA
(at 100)
mAA
(at 100)
Authors Keypoint Descriptor Summary Paper Website Processing date Flags
Submission ID: 00710
DISK-32D (epi)
Size: 128 bytes. Matches: built-in
2048.00 352.43
Rank: 1/6
0.459
Rank: 1/6
0.826
Rank: 2/6
0.45946
(±0.00023)
Rank: 2/6
484.30
Rank: 1/6
2324.68
Rank: 1/6
5.589
Rank: 1/6
0.427
Rank: 2/6
0.69938
(±0.00192)
Rank: 2/6
0.57942
Rank: 2/6
Under review! NeurIPS anonymous submission, ID 1194 (To be released: 21-09-16) disk-2020-09-15-nms-5-epi-32-save-49-imsize-1024-nump-2048 disk-2020-09-15-nms-5-epi-32-save-49-imsize-1024-nump-2048 (32 float32: 128 bytes) Local feature model learned via policy gradient, using 32D descriptors. Model trained with a cycle-consistency loss and supervised with epipolar constraints. Trained on MegaDepth, removing conflicts with the test data. For inference, images are resized to 1024 pixels on the longest edge, with NMS over a 5x5 window, taking the top 2048 features by score. N/A N/A 20-09-16 is_baseline, is_under_review
Submission ID: 00017
ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
2031.83 63.43
Rank: 6/6
0.366
Rank: 3/6
0.523
Rank: 4/6
0.08243
(±0.00058)
Rank: 4/6
120.93
Rank: 6/6
280.38
Rank: 6/6
2.191
Rank: 6/6
0.863
Rank: 4/6
0.02318
(±0.00229)
Rank: 6/6
0.05280
Rank: 6/6
Challenge organizers (contact) orb orb (32 uint8: 32 bytes) ORB with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00026
AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
1657.83 69.77
Rank: 5/6
0.355
Rank: 4/6
0.622
Rank: 3/6
0.17632
(±0.00087)
Rank: 3/6
224.40
Rank: 5/6
795.54
Rank: 5/6
3.311
Rank: 3/6
0.819
Rank: 3/6
0.23020
(±0.00466)
Rank: 3/6
0.20326
Rank: 3/6
Challenge organizers (contact) akaze-def akaze (61 uint8: 61 bytes) AKAZE with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00027
DELF-GLD (32D), PyRANSAC
Size: 128 bytes. Matches: built-in
2036.82 91.18
Rank: 4/6
0.109
Rank: 5/6
0.122
Rank: 5/6
0.06197
(±0.00032)
Rank: 6/6
453.19
Rank: 2/6
1805.48
Rank: 3/6
2.425
Rank: 4/6
0.916
Rank: 5/6
0.09163
(±0.00140)
Rank: 4/6
0.07680
Rank: 5/6
Challenge organizers (contact) delf-gld-2k-32d delf-gld-2k-32d (32 float32: 128 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 32 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with PyRANSAC (DEGENSAC with the degeneracy check turned off). https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-22 is_baseline
Submission ID: 00025
DELF-GLD (32D), DEGENSAC
Size: 128 bytes. Matches: built-in
2036.82 96.56
Rank: 3/6
0.109
Rank: 5/6
0.119
Rank: 6/6
0.06573
(±0.00030)
Rank: 5/6
453.19
Rank: 2/6
1805.48
Rank: 3/6
2.425
Rank: 4/6
0.920
Rank: 6/6
0.09163
(±0.00140)
Rank: 4/6
0.07868
Rank: 4/6
Challenge organizers (contact) delf-gld-2k-32d delf-gld-2k-32d (32 float32: 128 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 32 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with DEGENSAC. https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-22 is_baseline
Submission ID: 00711
DISK-32D (depth)
Size: 128 bytes. Matches: built-in
2048.00 341.83
Rank: 2/6
0.448
Rank: 2/6
0.842
Rank: 1/6
0.47005
(±0.00037)
Rank: 1/6
441.16
Rank: 4/6
2231.55
Rank: 2/6
5.524
Rank: 2/6
0.415
Rank: 1/6
0.70220
(±0.00208)
Rank: 1/6
0.58613
Rank: 1/6
Under review! NeurIPS anonymous submission, ID 1194 (To be released: 21-09-16) disk-2020-09-15-nms-7-depth-32-save-46-imsize-1024-nump-2048 disk-2020-09-15-nms-7-depth-32-save-46-imsize-1024-nump-2048 (32 float32: 128 bytes) Local feature model learned via policy gradient, using 32D descriptors. Model trained with a cycle-consistency loss and supervised with depth-based constraints. Trained on MegaDepth, removing conflicts with the test data. For inference, images are resized to 1024 pixels on the longest edge, with NMS over a 7x7 window, taking the top 2048 features by score. N/A N/A 20-09-16 is_baseline, is_under_review

Phototourism: unlimited keypoints, large descriptors (2048 bytes)

Note: entries with the same multi-view configuration may seem duplicated. This is normal: performance is averaged across tasks.
Stereo Multiview Avg.
Method NF NI Rep.
(3 pix.)
MS
(3 pix.)
mAA
(at 10o)
NM NL TL ATE mAA
(at 100)
mAA
(at 100)
Authors Keypoint Descriptor Summary Paper Website Processing date Flags
Submission ID: 00078
SOSNet-Upright w/ MAGSAC (no FLA...
Size: 512 bytes. Matches: built-in
7829.63 679.94
Rank: 17/130
0.486
Rank: 44/130
0.855
Rank: 30/130
0.56626
(±0.00020)
Rank: 23/130
600.23
Rank: 36/130
4765.43
Rank: 27/130
4.584
Rank: 27/130
0.403
Rank: 26/130
0.73027
(±0.00284)
Rank: 25/130
0.64826
Rank: 26/130
Challenge organizers (contact) sift8k sosnet-upright (128 float32: 512 bytes) Upright SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00597
SIFT8k-HardNet64
Size: 512 bytes. Matches: built-in
6589.88 381.75
Rank: 83/130
0.467
Rank: 108/130
0.852
Rank: 36/130
0.52511
(±0.00056)
Rank: 52/130
420.42
Rank: 78/130
3595.93
Rank: 82/130
4.496
Rank: 47/130
0.417
Rank: 42/130
0.70738
(±0.00218)
Rank: 48/130
0.61625
Rank: 45/130
caoliang (contact) sift8k hardnet (128 float32: 512 bytes) SIFT up to 8000 keypoints, hardnet extract descriptors.Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-05-27 is_submission, is_challenge_2020
Submission ID: 00042
ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
7150.21 161.98
Rank: 126/130
0.514
Rank: 16/130
0.653
Rank: 115/130
0.16159
(±0.00090)
Rank: 121/130
910.31
Rank: 13/130
1423.38
Rank: 128/130
2.722
Rank: 128/130
0.897
Rank: 130/130
0.08054
(±0.00242)
Rank: 130/130
0.12106
Rank: 130/130
Challenge organizers (contact) orb orb (32 uint8: 32 bytes) ORB with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00112
VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 204.37
Rank: 122/130
0.547
Rank: 12/130
0.823
Rank: 87/130
0.36954
(±0.00081)
Rank: 110/130
347.39
Rank: 111/130
3209.10
Rank: 103/130
4.126
Rank: 105/130
0.517
Rank: 114/130
0.58657
(±0.00120)
Rank: 113/130
0.47806
Rank: 112/130
Challenge organizers (contact) hessian vlsift (128 float32: 512 bytes) VL-Hess-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00508
modified-r2d2
Size: 512 bytes. Matches: built-in
7791.82 160.64
Rank: 127/130
0.517
Rank: 15/130
0.785
Rank: 103/130
0.32815
(±0.00037)
Rank: 111/130
328.39
Rank: 117/130
2811.27
Rank: 115/130
4.250
Rank: 91/130
0.488
Rank: 106/130
0.62899
(±0.00193)
Rank: 102/130
0.47857
Rank: 111/130
caoliang (contact) modified-r2d2 modified-r2d2 (128 float32: 512 bytes) r2d2 with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value to increase performance. https://europe.naverlabs.com/wp-content/uploads/2019/09/R2D2-Repeatable-and-Reliable-Detector-and-Descriptor-2.pdf https://github.com/naver/r2d2 20-04-25 is_submission, is_challenge_2020
Submission ID: 00536
HarrisZ + DeepOrientation + SOSN...
Size: 512 bytes. Matches: custom
2410.18 1164.84
Rank: 3/130
0.368
Rank: 119/130
0.341
Rank: 128/130
0.00607
(±0.00000)
Rank: 129/130
1191.31
Rank: 4/130
2215.65
Rank: 120/130
4.014
Rank: 116/130
0.524
Rank: 116/130
0.58741
(±0.00250)
Rank: 112/130
0.29674
Rank: 122/130
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] N/A N/A 20-05-01 is_submission, is_challenge_2020
Submission ID: 00019
Upright Root-SIFT (OpenCV), DEGE...
Size: 512 bytes. Matches: built-in
7829.24 355.69
Rank: 93/130
0.487
Rank: 23/130
0.844
Rank: 55/130
0.51009
(±0.00033)
Rank: 69/130
548.18
Rank: 43/130
4406.41
Rank: 47/130
4.385
Rank: 59/130
0.442
Rank: 70/130
0.68859
(±0.00254)
Rank: 72/130
0.59934
Rank: 68/130
Challenge organizers (contact) sift-lowth rootsift-upright (128 float32: 512 bytes) Upright Root-SIFT with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00085
LogPolar-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 611.19
Rank: 26/130
0.486
Rank: 44/130
0.830
Rank: 75/130
0.51287
(±0.00038)
Rank: 66/130
483.93
Rank: 63/130
4405.57
Rank: 48/130
4.542
Rank: 39/130
0.423
Rank: 52/130
0.70680
(±0.00247)
Rank: 49/130
0.60983
Rank: 60/130
Challenge organizers (contact) sift8k logpolar96-fixed-upright (128 float32: 512 bytes) Upright LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-23 is_baseline
Submission ID: 00571
ContextDesc Upright + OANetV2 + ...
Size: 512 bytes. Matches: custom
7830.09 827.45
Rank: 9/130
0.487
Rank: 25/130
0.775
Rank: 104/130
0.53810
(±0.00000)
Rank: 43/130
855.38
Rank: 16/130
6561.52
Rank: 3/130
4.522
Rank: 41/130
0.370
Rank: 8/130
0.76702
(±0.00081)
Rank: 6/130
0.65256
Rank: 23/130
Jiahui Zhang, Zixin Luo, Hongkai Chen (contact) contextdesc-upright contextdesc-upright (128 float32: 512 bytes) ContextDesc with 8000 SIFT features, using improved OANet matcher and DEGENSAC post-processing N/A N/A 20-05-21 is_submission, is_challenge_2020
Submission ID: 00709
DISK (LCC/depth)
Size: 512 bytes. Matches: built-in
7844.17 1238.52
Rank: 2/130
0.644
Rank: 1/130
0.852
Rank: 37/130
0.55847
(±0.00084)
Rank: 27/130
1663.81
Rank: 1/130
7483.98
Rank: 2/130
5.922
Rank: 1/130
0.391
Rank: 23/130
0.75024
(±0.00316)
Rank: 15/130
0.65435
Rank: 21/130
Under review! NeurIPS anonymous submission, ID 1194 (To be released: 21-06-02) disk-cc-continued-20-imsize-1024-nms-3-nump-8000 disk-cc-continued-20-imsize-1024-nms-3-nump-8000 (128 float32: 512 bytes) Local feature model learned via policy gradient. Model trained with a cycle-consistency loss and supervised with depth. Trained on MegaDepth, removing conflicts with the test data. For inference, images are resized to 1024 pixels on the longest edge, with NMS over a 3x3 window. We take the top 8000 features by score. N/A N/A 20-06-02 is_submission, is_challenge_2020, is_under_review
Submission ID: 00573
Guided-HardNet-qht
Size: 512 bytes. Matches: custom
7829.63 887.83
Rank: 7/130
0.486
Rank: 44/130
0.774
Rank: 105/130
0.55224
(±0.00000)
Rank: 31/130
538.28
Rank: 45/130
4693.85
Rank: 30/130
4.650
Rank: 9/130
0.381
Rank: 11/130
0.75930
(±0.00177)
Rank: 8/130
0.65577
Rank: 20/130
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet-qht (128 float32: 512 bytes) sift and hardnet with 8k features, using the modified guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-23 is_submission, is_challenge_2020
Submission ID: 00530
SIFT-patch
Size: 1024 bytes. Matches: built-in
7861.62 83.83
Rank: 130/130
0.472
Rank: 83/130
0.577
Rank: 120/130
0.22977
(±0.00029)
Rank: 117/130
80.10
Rank: 130/130
973.09
Rank: 130/130
2.007
Rank: 129/130
0.726
Rank: 127/130
0.08184
(±0.00372)
Rank: 129/130
0.15581
Rank: 129/130
feyman_priv (contact) sift8k l2net-ith-rcface (256 float32: 1024 bytes) sift8k with l2net trained on google-landmark-dataset-v1(in 1000 class) N/A N/A 20-04-24 is_submission, is_challenge_2020
Submission ID: 00010
AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
7857.11 246.74
Rank: 117/130
0.553
Rank: 9/130
0.735
Rank: 111/130
0.30717
(±0.00122)
Rank: 114/130
479.55
Rank: 67/130
2778.68
Rank: 116/130
3.393
Rank: 125/130
0.737
Rank: 129/130
0.36048
(±0.00382)
Rank: 127/130
0.33383
Rank: 116/130
Challenge organizers (contact) akaze-lowth akaze (61 uint8: 61 bytes) AKAZE with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00068
GeoDesc-Upright w/ DEGENSAC (no ...
Size: 512 bytes. Matches: built-in
7829.63 409.87
Rank: 75/130
0.486
Rank: 44/130
0.869
Rank: 17/130
0.52674
(±0.00049)
Rank: 50/130
458.60
Rank: 68/130
4146.83
Rank: 59/130
4.412
Rank: 55/130
0.431
Rank: 61/130
0.70442
(±0.00189)
Rank: 53/130
0.61558
Rank: 46/130
Challenge organizers (contact) sift8k geodesc-upright (128 float32: 512 bytes) Upright GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00532
affnet-hardnet
Size: 512 bytes. Matches: built-in
7925.53 226.75
Rank: 119/130
0.452
Rank: 114/130
0.649
Rank: 117/130
0.31958
(±0.00055)
Rank: 112/130
243.28
Rank: 126/130
3155.72
Rank: 106/130
3.595
Rank: 121/130
0.554
Rank: 117/130
0.52578
(±0.00343)
Rank: 117/130
0.42268
Rank: 114/130
caoliang (contact) affnet hardnet (128 float32: 512 bytes) affnet up to 8000 keypoints, harnet extract descriptors.Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo https://arxiv.org/abs/1711.06704 https://github.com/ducha-aiki/affnet 20-04-24 is_submission, is_challenge_2020
Submission ID: 00071
L2-Net w/ DEGENSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 366.00
Rank: 88/130
0.472
Rank: 87/130
0.852
Rank: 40/130
0.52953
(±0.00043)
Rank: 49/130
339.00
Rank: 115/130
3424.94
Rank: 92/130
4.206
Rank: 97/130
0.452
Rank: 85/130
0.66445
(±0.00213)
Rank: 87/130
0.59699
Rank: 71/130
Challenge organizers (contact) sift8k l2net (128 float32: 512 bytes) L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-23 is_baseline
Submission ID: 00024
SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
7860.73 238.66
Rank: 118/130
0.472
Rank: 81/130
0.824
Rank: 85/130
0.45426
(±0.00097)
Rank: 99/130
418.86
Rank: 80/130
3515.63
Rank: 89/130
4.001
Rank: 117/130
0.502
Rank: 109/130
0.60193
(±0.00185)
Rank: 110/130
0.52810
Rank: 102/130
Challenge organizers (contact) sift-lowth sift (128 float32: 512 bytes) SIFT with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00084
LogPolar-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 459.75
Rank: 61/130
0.486
Rank: 44/130
0.852
Rank: 38/130
0.52316
(±0.00034)
Rank: 57/130
483.93
Rank: 63/130
4405.57
Rank: 48/130
4.542
Rank: 39/130
0.422
Rank: 49/130
0.70680
(±0.00247)
Rank: 49/130
0.61498
Rank: 48/130
Challenge organizers (contact) sift8k logpolar96-fixed-upright (128 float32: 512 bytes) Upright LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00506
sift and hardnet64
Size: 512 bytes. Matches: built-in
7861.62 358.55
Rank: 91/130
0.472
Rank: 85/130
0.849
Rank: 43/130
0.52085
(±0.00009)
Rank: 58/130
585.33
Rank: 38/130
4957.86
Rank: 25/130
4.294
Rank: 83/130
0.419
Rank: 46/130
0.70472
(±0.00257)
Rank: 52/130
0.61279
Rank: 53/130
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64 (128 float32: 512 bytes) SIFT with 8000 keypoints, hardnet64 with 128 descriptors N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00100
KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 448.11
Rank: 64/130
0.582
Rank: 2/130
0.838
Rank: 62/130
0.39971
(±0.00147)
Rank: 105/130
356.21
Rank: 102/130
3366.01
Rank: 95/130
4.319
Rank: 77/130
0.462
Rank: 94/130
0.64829
(±0.00174)
Rank: 95/130
0.52400
Rank: 103/130
Challenge organizers (contact) keynettuned vlhardnet (128 float32: 512 bytes) KeyNet-HardNet with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00569
Upright-SIFT-HardNet
Size: 512 bytes. Matches: built-in
6667.86 376.15
Rank: 85/130
0.491
Rank: 18/130
0.844
Rank: 53/130
0.51560
(±0.00043)
Rank: 63/130
397.09
Rank: 89/130
3417.56
Rank: 94/130
4.479
Rank: 50/130
0.448
Rank: 84/130
0.68803
(±0.00532)
Rank: 73/130
0.60181
Rank: 66/130
caoliang (contact) sift8k hardnet (128 float32: 512 bytes) SIFT up to 8000 keypoints, harnet extract descriptors.Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-05-20 is_submission, is_challenge_2020
Submission ID: 00066
SOSNet-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 537.85
Rank: 43/130
0.486
Rank: 44/130
0.841
Rank: 58/130
0.53531
(±0.00061)
Rank: 45/130
521.37
Rank: 53/130
4473.48
Rank: 42/130
4.597
Rank: 23/130
0.414
Rank: 38/130
0.72179
(±0.00330)
Rank: 34/130
0.62855
Rank: 40/130
Challenge organizers (contact) sift8k sosnet-upright (128 float32: 512 bytes) Upright SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00076
SOSNet w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 563.27
Rank: 41/130
0.472
Rank: 87/130
0.846
Rank: 47/130
0.55170
(±0.00042)
Rank: 32/130
508.66
Rank: 58/130
4502.19
Rank: 38/130
4.377
Rank: 68/130
0.405
Rank: 32/130
0.71820
(±0.00298)
Rank: 39/130
0.63495
Rank: 34/130
Challenge organizers (contact) sift8k sosnet (128 float32: 512 bytes) SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00065
SOSNet-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 407.29
Rank: 76/130
0.486
Rank: 44/130
0.864
Rank: 26/130
0.54098
(±0.00086)
Rank: 39/130
521.37
Rank: 53/130
4473.48
Rank: 42/130
4.597
Rank: 23/130
0.417
Rank: 40/130
0.72179
(±0.00330)
Rank: 34/130
0.63139
Rank: 36/130
Challenge organizers (contact) sift8k sosnet-upright (128 float32: 512 bytes) Upright SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00522
SIFT and HardNet64 train scale(5...
Size: 512 bytes. Matches: built-in
7830.09 513.57
Rank: 50/130
0.486
Rank: 32/130
0.852
Rank: 41/130
0.55385
(±0.00076)
Rank: 30/130
822.07
Rank: 18/130
5660.19
Rank: 16/130
4.555
Rank: 34/130
0.404
Rank: 30/130
0.72744
(±0.00455)
Rank: 28/130
0.64065
Rank: 31/130
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-raw64 (128 float32: 512 bytes) SIFT with 8000 keypoints(raw 64), hardnet64 with 128 descriptors, FLANN disabled N/A N/A 20-04-26 is_submission, is_challenge_2020
Submission ID: 00554
ContextDesc-Upright w/ DEGENSAC ...
Size: 512 bytes. Matches: built-in
7830.09 470.69
Rank: 59/130
0.487
Rank: 25/130
0.877
Rank: 3/130
0.55697
(±0.00079)
Rank: 28/130
543.45
Rank: 44/130
4543.76
Rank: 37/130
4.498
Rank: 46/130
0.405
Rank: 31/130
0.72902
(±0.00255)
Rank: 27/130
0.64299
Rank: 30/130
Zixin Luo, Jiahui Zhang, Hongkai Chen (contact) sift-def contextdesc-upright (128 float32: 512 bytes) ContextDesc with 8000 SIFT features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value and use only single orientation to increase performance. FLANN disabled. https://arxiv.org/abs/1904.04084 https://github.com/lzx551402/contextdesc 20-05-11 is_submission, is_challenge_2020
Submission ID: 00504
ASLFeat (MS) w/ DEGENSAC (no FLA...
Size: 512 bytes. Matches: built-in
6948.83 390.79
Rank: 79/130
0.550
Rank: 10/130
0.797
Rank: 100/130
0.46102
(±0.00067)
Rank: 96/130
510.80
Rank: 55/130
3817.14
Rank: 75/130
4.577
Rank: 29/130
0.444
Rank: 75/130
0.68246
(±0.00173)
Rank: 76/130
0.57174
Rank: 88/130
Zixin Luo, Jiahui Zhang (contact) aslfeat aslfeat (128 float32: 512 bytes) ASLFeat (MS) with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC. FLANN disabled. https://arxiv.org/abs/2003.10071 https://github.com/lzx551402/ASLFeat 20-04-25 is_submission, is_challenge_2020
Submission ID: 00548
HarrisZ + DeepOrientation + SOSN...
Size: 512 bytes. Matches: custom
2410.18 343.78
Rank: 97/130
0.368
Rank: 119/130
0.607
Rank: 119/130
0.28795
(±0.00000)
Rank: 115/130
351.53
Rank: 105/130
1699.08
Rank: 125/130
4.124
Rank: 108/130
0.507
Rank: 110/130
0.61370
(±0.00246)
Rank: 106/130
0.45082
Rank: 113/130
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 0.75, degeneracy check true) [Mishkin 2019] N/A N/A 20-05-03 is_submission, is_challenge_2020
Submission ID: 00519
SOSNet-Upright-AdaLAM
Size: 512 bytes. Matches: custom
7830.09 803.50
Rank: 11/130
0.486
Rank: 32/130
0.818
Rank: 92/130
0.56056
(±0.00000)
Rank: 25/130
827.85
Rank: 17/130
6157.62
Rank: 10/130
4.700
Rank: 5/130
0.381
Rank: 12/130
0.75917
(±0.00446)
Rank: 9/130
0.65987
Rank: 18/130
Luca Cavalli, Viktor Larsson, Martin Oswald, Torsten Sattler, Marc Pollefeys (contact) sift-def sosnet-upright (128 float32: 512 bytes) Using upright SOSNet descriptors with 8000 features, nearest neighbor matching and outlier rejection enforcing local affine consistency within a confidence-based adaptive error tolerance. Matches post-processed with DEGENSAC. https://arxiv.org/abs/2006.04250 N/A 20-04-26 is_submission, is_challenge_2020
Submission ID: 00101
KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 598.28
Rank: 32/130
0.582
Rank: 2/130
0.826
Rank: 82/130
0.49856
(±0.00038)
Rank: 75/130
356.21
Rank: 102/130
3366.01
Rank: 95/130
4.319
Rank: 77/130
0.464
Rank: 97/130
0.64829
(±0.00174)
Rank: 95/130
0.57342
Rank: 86/130
Challenge organizers (contact) keynettuned vlhardnet (128 float32: 512 bytes) KeyNet-HardNet with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00501
Guided matching with Upright roo...
Size: 512 bytes. Matches: custom
7829.27 609.35
Rank: 27/130
0.486
Rank: 29/130
0.786
Rank: 102/130
0.49072
(±0.00000)
Rank: 79/130
1363.98
Rank: 3/130
7835.75
Rank: 1/130
4.219
Rank: 96/130
0.454
Rank: 87/130
0.68064
(±0.00390)
Rank: 79/130
0.58568
Rank: 77/130
(contact) sift upright-root-sift (128 float32: 512 bytes) In submission N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00060
L2-Net w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 386.81
Rank: 80/130
0.472
Rank: 87/130
0.813
Rank: 96/130
0.49054
(±0.00043)
Rank: 80/130
314.54
Rank: 121/130
3314.19
Rank: 101/130
4.204
Rank: 99/130
0.465
Rank: 98/130
0.65665
(±0.00287)
Rank: 89/130
0.57360
Rank: 85/130
Challenge organizers (contact) sift8k l2net (128 float32: 512 bytes) L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-22 is_baseline
Submission ID: 00545
structured feature description
Size: 512 bytes. Matches: built-in
7830.21 411.68
Rank: 73/130
0.486
Rank: 42/130
0.865
Rank: 23/130
0.54442
(±0.00084)
Rank: 37/130
524.05
Rank: 51/130
4467.23
Rank: 44/130
4.551
Rank: 35/130
0.412
Rank: 35/130
0.72729
(±0.00300)
Rank: 29/130
0.63586
Rank: 33/130
Mahdi Abolfazli Esfahani, Han Wang (contact) sift sfd (128 float32: 512 bytes) structured descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC N/A N/A 20-04-29 is_submission, is_challenge_2020
Submission ID: 00120
CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 265.75
Rank: 113/130
0.472
Rank: 78/130
0.873
Rank: 11/130
0.46073
(±0.00062)
Rank: 97/130
356.56
Rank: 99/130
3550.63
Rank: 84/130
4.275
Rank: 85/130
0.439
Rank: 67/130
0.68875
(±0.00138)
Rank: 69/130
0.57474
Rank: 84/130
Challenge organizers (contact) sift8k hardnetamos (128 float32: 512 bytes) CV-DoG-HardNetAmos with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00568
Guided-SOSNet-lib-p
Size: 512 bytes. Matches: custom
7829.63 508.45
Rank: 51/130
0.486
Rank: 44/130
0.874
Rank: 10/130
0.57982
(±0.00000)
Rank: 16/130
524.66
Rank: 50/130
4618.86
Rank: 33/130
4.632
Rank: 16/130
0.369
Rank: 6/130
0.75888
(±0.00437)
Rank: 10/130
0.66935
Rank: 9/130
Chen Shen, Zhipeng Wang, Jun Zhang, Zhiwei Ruan, Zhongkun Chen, Jingchao Zhou, Pengfei Xu (contact) sift8k sosnet-lib-p (128 float32: 512 bytes) sift and sosnet with 8k features, using the guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-15 is_submission, is_challenge_2020
Submission ID: 00051
HardNet w/ DEGENSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 432.32
Rank: 70/130
0.472
Rank: 87/130
0.866
Rank: 22/130
0.55430
(±0.00031)
Rank: 29/130
426.84
Rank: 76/130
4001.40
Rank: 67/130
4.339
Rank: 75/130
0.419
Rank: 44/130
0.70962
(±0.00191)
Rank: 46/130
0.63196
Rank: 35/130
Challenge organizers (contact) sift8k hardnet (128 float32: 512 bytes) HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00518
UprightRootSIFT-AdaLAM
Size: 512 bytes. Matches: custom
6449.42 435.53
Rank: 69/130
0.436
Rank: 116/130
0.825
Rank: 84/130
0.49625
(±0.00000)
Rank: 76/130
449.49
Rank: 71/130
4353.56
Rank: 50/130
4.453
Rank: 53/130
0.397
Rank: 24/130
0.72490
(±0.00190)
Rank: 30/130
0.61057
Rank: 57/130
Luca Cavalli, Viktor Larsson, Martin Oswald, Torsten Sattler, Marc Pollefeys (contact) sift-def rootsift-upright (128 float32: 512 bytes) Using upright RootSIFT with 8000 features, nearest neighbor matching and outlier rejection enforcing local affine consistency within a confidence-based adaptive error tolerance. Matches post-processed with DEGENSAC. https://arxiv.org/abs/2006.04250 N/A 20-04-24 is_submission, is_challenge_2020
Submission ID: 00058
HardNet-Upright w/ MAGSAC (no FL...
Size: 512 bytes. Matches: built-in
7829.63 707.86
Rank: 16/130
0.486
Rank: 44/130
0.853
Rank: 32/130
0.56374
(±0.00069)
Rank: 24/130
509.07
Rank: 56/130
4250.40
Rank: 51/130
4.548
Rank: 36/130
0.412
Rank: 36/130
0.72309
(±0.00141)
Rank: 32/130
0.64341
Rank: 29/130
Challenge organizers (contact) sift8k hardnet-upright (128 float32: 512 bytes) Upright HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00034
D2-Net (single-scale), DEGENSAC
Size: 2048 bytes. Matches: built-in
5665.31 251.55
Rank: 115/130
0.288
Rank: 125/130
0.360
Rank: 127/130
0.18499
(±0.00022)
Rank: 120/130
740.01
Rank: 26/130
4584.07
Rank: 36/130
3.506
Rank: 122/130
0.696
Rank: 126/130
0.41726
(±0.00175)
Rank: 125/130
0.30112
Rank: 121/130
Challenge organizers (contact) d2net-singlescale d2net-singlescale (512 float32: 2048 bytes) D2-Net, single-scale model, up to 8000 features. Trained on the MegaDepth dataset, removing scenes which overlap with the Phototourism test set. Stereo with DEGENSAC and optimal parameters http://openaccess.thecvf.com/content_CVPR_2019/papers/Dusmanu_D2-Net_A_Trainable_CNN_for_Joint_Description_and_Detection_of_CVPR_2019_paper.pdf https://github.com/mihaidusmanu/d2-net 20-04-22 is_baseline
Submission ID: 00122
CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 528.67
Rank: 45/130
0.472
Rank: 78/130
0.838
Rank: 61/130
0.53291
(±0.00047)
Rank: 47/130
356.56
Rank: 99/130
3550.63
Rank: 84/130
4.275
Rank: 85/130
0.443
Rank: 74/130
0.68875
(±0.00138)
Rank: 69/130
0.61083
Rank: 56/130
Challenge organizers (contact) sift8k hardnetamos (128 float32: 512 bytes) CV-DoG-HardNetAmos with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00061
L2-Net-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 358.37
Rank: 92/130
0.486
Rank: 44/130
0.845
Rank: 49/130
0.52012
(±0.00106)
Rank: 60/130
369.60
Rank: 97/130
3538.91
Rank: 87/130
4.407
Rank: 57/130
0.447
Rank: 81/130
0.68108
(±0.00177)
Rank: 77/130
0.60060
Rank: 67/130
Challenge organizers (contact) sift8k l2net-upright (128 float32: 512 bytes) Upright L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-23 is_baseline
Submission ID: 00121
CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 398.65
Rank: 78/130
0.472
Rank: 78/130
0.863
Rank: 27/130
0.53852
(±0.00039)
Rank: 42/130
356.56
Rank: 99/130
3550.63
Rank: 84/130
4.275
Rank: 85/130
0.439
Rank: 66/130
0.68875
(±0.00138)
Rank: 69/130
0.61364
Rank: 50/130
Challenge organizers (contact) sift8k hardnetamos (128 float32: 512 bytes) CV-DoG-HardNetAmos with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00069
GeoDesc w/ DEGENSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 348.52
Rank: 96/130
0.472
Rank: 87/130
0.856
Rank: 29/130
0.51112
(±0.00070)
Rank: 68/130
395.12
Rank: 92/130
3838.97
Rank: 73/130
4.264
Rank: 88/130
0.443
Rank: 72/130
0.68032
(±0.00099)
Rank: 80/130
0.59572
Rank: 73/130
Challenge organizers (contact) sift8k geodesc (128 float32: 512 bytes) GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00625
HardNet64-train-all-l2-val-14000...
Size: 512 bytes. Matches: built-in
7830.09 605.12
Rank: 29/130
0.486
Rank: 32/130
0.872
Rank: 14/130
0.58776
(±0.00048)
Rank: 7/130
899.14
Rank: 14/130
6095.78
Rank: 11/130
4.645
Rank: 11/130
0.389
Rank: 19/130
0.74849
(±0.00117)
Rank: 17/130
0.66812
Rank: 11/130
Ximin Zheng, Sheng He, Guanlin Liang (contact) sift8k hardnet64-train-all-l2-val-14000 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 14000), FLANN disabled N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00591
Guided-Hardnet-Epoch2
Size: 512 bytes. Matches: custom
7829.63 761.99
Rank: 13/130
0.486
Rank: 44/130
0.823
Rank: 88/130
0.61081
(±0.00000)
Rank: 1/130
785.57
Rank: 23/130
6330.70
Rank: 7/130
4.680
Rank: 7/130
0.358
Rank: 2/130
0.78288
(±0.00094)
Rank: 2/130
0.69684
Rank: 1/130
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet-epoch2 (128 float32: 512 bytes) sift and hardnet with 8k features, using the modified guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-24 is_submission, is_challenge_2020
Submission ID: 00086
LogPolar-Upright w/ MAGSAC (no F...
Size: 512 bytes. Matches: built-in
7829.63 732.28
Rank: 15/130
0.486
Rank: 44/130
0.844
Rank: 54/130
0.54060
(±0.00047)
Rank: 40/130
505.37
Rank: 61/130
4414.11
Rank: 45/130
4.518
Rank: 43/130
0.421
Rank: 48/130
0.71092
(±0.00251)
Rank: 43/130
0.62576
Rank: 43/130
Challenge organizers (contact) sift8k logpolar96-fixed-upright (128 float32: 512 bytes) Upright LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00106
VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 326.20
Rank: 103/130
0.490
Rank: 20/130
0.809
Rank: 97/130
0.46326
(±0.00089)
Rank: 94/130
324.62
Rank: 118/130
3030.67
Rank: 110/130
4.173
Rank: 101/130
0.462
Rank: 95/130
0.62829
(±0.00112)
Rank: 103/130
0.54578
Rank: 97/130
Challenge organizers (contact) dog vlsift (128 float32: 512 bytes) VL-DoG-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00081
LogPolar w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 591.21
Rank: 33/130
0.472
Rank: 87/130
0.832
Rank: 71/130
0.52385
(±0.00113)
Rank: 55/130
415.77
Rank: 82/130
4054.60
Rank: 62/130
4.316
Rank: 80/130
0.432
Rank: 62/130
0.69284
(±0.00429)
Rank: 63/130
0.60835
Rank: 61/130
Challenge organizers (contact) sift8k logpolar96-fixed (128 float32: 512 bytes) LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00562
HardNet64-train-all-raw64-balanc...
Size: 512 bytes. Matches: built-in
7831.92 370.11
Rank: 87/130
0.486
Rank: 73/130
0.807
Rank: 98/130
0.45770
(±0.00060)
Rank: 98/130
575.43
Rank: 41/130
4756.46
Rank: 29/130
4.563
Rank: 33/130
0.435
Rank: 65/130
0.69780
(±0.00191)
Rank: 58/130
0.57775
Rank: 81/130
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-raw64-balance-l2-224000-raw64 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 224000), FLANN disabled N/A N/A 20-05-13 is_submission, is_challenge_2020
Submission ID: 00557
HarrisZ + DeepOrientation + SOSN...
Size: 512 bytes. Matches: custom
2410.18 614.16
Rank: 24/130
0.368
Rank: 119/130
0.513
Rank: 122/130
0.05482
(±0.00000)
Rank: 124/130
626.57
Rank: 32/130
1793.00
Rank: 124/130
4.314
Rank: 82/130
0.517
Rank: 113/130
0.56933
(±0.00268)
Rank: 116/130
0.31207
Rank: 119/130
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 10) [Mishkin 2019] N/A N/A 20-05-11 is_submission, is_challenge_2020
Submission ID: 00621
CV-DoG-HardNet8-PTv2
Size: 512 bytes. Matches: built-in
7830.09 583.12
Rank: 35/130
0.486
Rank: 31/130
0.880
Rank: 2/130
0.58997
(±0.00054)
Rank: 4/130
577.37
Rank: 39/130
4476.61
Rank: 41/130
4.638
Rank: 15/130
0.407
Rank: 33/130
0.73096
(±0.00227)
Rank: 24/130
0.66046
Rank: 17/130
Milan Pultar, Dmytro Mishkin, Jiri Matas (contact) sift8k h8e512pt (128 float32: 512 bytes) HardNet8 with PCA compression, batch sampling from few images N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00520
SIFT and HardNet64 train scale(5...
Size: 512 bytes. Matches: built-in
7830.09 400.19
Rank: 77/130
0.486
Rank: 32/130
0.837
Rank: 66/130
0.50589
(±0.00059)
Rank: 70/130
619.25
Rank: 34/130
4959.78
Rank: 24/130
4.384
Rank: 60/130
0.456
Rank: 89/130
0.69172
(±0.00048)
Rank: 65/130
0.59880
Rank: 69/130
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-scale5 (128 float32: 512 bytes) SIFT with 8000 keypoints(size scaled by 5), hardnet64 with 128 descriptors, FLANN disabled N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00526
ASLFeat-MS
Size: 512 bytes. Matches: built-in
7384.19 456.55
Rank: 62/130
0.578
Rank: 5/130
0.752
Rank: 110/130
0.42208
(±0.00054)
Rank: 102/130
532.83
Rank: 48/130
4002.80
Rank: 66/130
4.521
Rank: 42/130
0.447
Rank: 78/130
0.66568
(±0.00347)
Rank: 85/130
0.54388
Rank: 99/130
Zixin Luo, Jiahui Zhang (contact) aslfeat-ms aslfeat-ms (128 float32: 512 bytes) ASLFeat (joint predition of detectors and descriptors with multi-scale input) with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC. N/A N/A 20-04-27 is_submission, is_challenge_2020
Submission ID: 00546
HarrisZ + DeepOrientation + SOSN...
Size: 512 bytes. Matches: custom
2410.18 1020.86
Rank: 6/130
0.368
Rank: 119/130
0.366
Rank: 126/130
0.07780
(±0.00000)
Rank: 123/130
1047.01
Rank: 7/130
2197.09
Rank: 121/130
4.023
Rank: 115/130
0.509
Rank: 111/130
0.59222
(±0.00276)
Rank: 111/130
0.33501
Rank: 115/130
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 15, degeneracy check true) [Mishkin 2019] N/A N/A 20-05-02 is_baseline
Submission ID: 00524
SIFT-patch
Size: 512 bytes. Matches: built-in
7861.62 84.08
Rank: 129/130
0.472
Rank: 83/130
0.574
Rank: 121/130
0.22805
(±0.00076)
Rank: 118/130
81.93
Rank: 129/130
991.42
Rank: 129/130
1.934
Rank: 130/130
0.687
Rank: 125/130
0.08662
(±0.00384)
Rank: 128/130
0.15733
Rank: 128/130
feyman_priv (contact) sift8k l2net-arcface (128 float32: 512 bytes) sift8k with l2net trained on google-landmark-dataset-v1(in 1000 class) N/A N/A 20-04-26 is_submission, is_challenge_2020
Submission ID: 00082
LogPolar w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 496.36
Rank: 53/130
0.472
Rank: 87/130
0.817
Rank: 93/130
0.49129
(±0.00011)
Rank: 78/130
399.70
Rank: 87/130
4036.37
Rank: 64/130
4.340
Rank: 73/130
0.429
Rank: 60/130
0.69100
(±0.00154)
Rank: 66/130
0.59115
Rank: 76/130
Challenge organizers (contact) sift8k logpolar96-fixed (128 float32: 512 bytes) LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00116
VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 209.26
Rank: 121/130
0.577
Rank: 6/130
0.841
Rank: 59/130
0.39330
(±0.00052)
Rank: 108/130
350.69
Rank: 106/130
3327.71
Rank: 98/130
4.076
Rank: 112/130
0.490
Rank: 108/130
0.60691
(±0.00338)
Rank: 107/130
0.50011
Rank: 110/130
Challenge organizers (contact) hessian affnetvlsift (128 float32: 512 bytes) VL-HessAffNet-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00611
sift and hardnet64 train scale(1...
Size: 512 bytes. Matches: built-in
7830.09 622.13
Rank: 22/130
0.486
Rank: 32/130
0.871
Rank: 15/130
0.58870
(±0.00041)
Rank: 5/130
899.14
Rank: 14/130
6086.16
Rank: 12/130
4.647
Rank: 10/130
0.386
Rank: 15/130
0.75127
(±0.00234)
Rank: 14/130
0.66999
Rank: 8/130
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-l2-val-14000 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 14000), FLANN disabled N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00063
SOSNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 340.09
Rank: 99/130
0.472
Rank: 87/130
0.854
Rank: 31/130
0.51891
(±0.00067)
Rank: 61/130
440.21
Rank: 73/130
4178.82
Rank: 54/130
4.378
Rank: 66/130
0.419
Rank: 45/130
0.70078
(±0.00375)
Rank: 56/130
0.60984
Rank: 59/130
Challenge organizers (contact) sift8k sosnet (128 float32: 512 bytes) SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00109
VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 250.11
Rank: 116/130
0.482
Rank: 75/130
0.838
Rank: 63/130
0.46795
(±0.00075)
Rank: 90/130
311.54
Rank: 123/130
3061.54
Rank: 107/130
4.105
Rank: 109/130
0.475
Rank: 101/130
0.62964
(±0.00428)
Rank: 99/130
0.54880
Rank: 94/130
Challenge organizers (contact) dogaffine vlsift (128 float32: 512 bytes) VL-DoGAff-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00118
VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 350.03
Rank: 94/130
0.577
Rank: 6/130
0.824
Rank: 86/130
0.46263
(±0.00079)
Rank: 95/130
350.69
Rank: 106/130
3327.71
Rank: 98/130
4.076
Rank: 112/130
0.489
Rank: 107/130
0.60691
(±0.00338)
Rank: 107/130
0.53477
Rank: 101/130
Challenge organizers (contact) hessian affnetvlsift (128 float32: 512 bytes) VL-HessAffNet-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00576
Upright-Sift + X-Net-lib w/ DEGE...
Size: 512 bytes. Matches: built-in
7830.16 465.31
Rank: 60/130
0.486
Rank: 43/130
0.844
Rank: 52/130
0.53156
(±0.00027)
Rank: 48/130
720.29
Rank: 27/130
5398.84
Rank: 20/130
4.547
Rank: 38/130
0.408
Rank: 34/130
0.71966
(±0.00160)
Rank: 38/130
0.62561
Rank: 44/130
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) sift x-net-lib (128 float32: 512 bytes) N/A N/A 20-05-20 is_submission, is_challenge_2020
Submission ID: 00075
SOSNet w/ DEGENSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 424.61
Rank: 72/130
0.472
Rank: 87/130
0.868
Rank: 19/130
0.55867
(±0.00060)
Rank: 26/130
508.66
Rank: 58/130
4502.19
Rank: 38/130
4.377
Rank: 68/130
0.404
Rank: 29/130
0.71820
(±0.00298)
Rank: 39/130
0.63844
Rank: 32/130
Challenge organizers (contact) sift8k sosnet (128 float32: 512 bytes) SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-23 is_baseline
Submission ID: 00105
VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 261.60
Rank: 114/130
0.490
Rank: 20/130
0.826
Rank: 83/130
0.46555
(±0.00055)
Rank: 93/130
324.62
Rank: 118/130
3030.67
Rank: 110/130
4.173
Rank: 101/130
0.461
Rank: 93/130
0.62829
(±0.00112)
Rank: 103/130
0.54692
Rank: 96/130
Challenge organizers (contact) dog vlsift (128 float32: 512 bytes) VL-DoG-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00062
L2-Net-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 471.61
Rank: 58/130
0.486
Rank: 44/130
0.822
Rank: 89/130
0.51450
(±0.00058)
Rank: 64/130
369.60
Rank: 97/130
3538.91
Rank: 87/130
4.407
Rank: 57/130
0.448
Rank: 83/130
0.68108
(±0.00177)
Rank: 77/130
0.59779
Rank: 70/130
Challenge organizers (contact) sift8k l2net-upright (128 float32: 512 bytes) Upright L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-22 is_baseline
Submission ID: 00048
GeoDesc w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 362.25
Rank: 90/130
0.472
Rank: 87/130
0.816
Rank: 94/130
0.46887
(±0.00025)
Rank: 89/130
350.43
Rank: 109/130
3601.67
Rank: 80/130
4.234
Rank: 92/130
0.447
Rank: 80/130
0.65553
(±0.00444)
Rank: 92/130
0.56220
Rank: 91/130
Challenge organizers (contact) sift8k geodesc (128 float32: 512 bytes) GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00056
HardNet-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 439.36
Rank: 67/130
0.486
Rank: 44/130
0.864
Rank: 25/130
0.54393
(±0.00027)
Rank: 38/130
480.13
Rank: 65/130
4167.34
Rank: 56/130
4.566
Rank: 31/130
0.417
Rank: 41/130
0.71542
(±0.00274)
Rank: 41/130
0.62967
Rank: 39/130
Challenge organizers (contact) sift8k hardnet-upright (128 float32: 512 bytes) Upright HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00613
HardNet64-data-aug-sort-51
Size: 512 bytes. Matches: built-in
7830.09 624.09
Rank: 21/130
0.486
Rank: 32/130
0.870
Rank: 16/130
0.58727
(±0.00076)
Rank: 8/130
964.80
Rank: 9/130
6350.68
Rank: 5/130
4.644
Rank: 14/130
0.383
Rank: 13/130
0.74952
(±0.00162)
Rank: 16/130
0.66839
Rank: 10/130
Ximin Zheng, Sheng He, Guanlin Liang (contact) sift8k hardnet64-data-aug-sort-51 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 124000 and data augument), FLANN disabled N/A N/A 20-05-31 is_submission, is_challenge_2020
Submission ID: 00598
HarrisZ (more kpt) + DeepOrienta...
Size: 512 bytes. Matches: custom
4478.94 563.62
Rank: 40/130
0.455
Rank: 112/130
0.711
Rank: 112/130
0.39767
(±0.00000)
Rank: 107/130
576.83
Rank: 40/130
2707.54
Rank: 117/130
4.381
Rank: 63/130
0.445
Rank: 77/130
0.69090
(±0.00163)
Rank: 68/130
0.54428
Rank: 98/130
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ (start scale 2) [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 0.5, degeneracy check true, confidence 0.98, max iter 500000) [Mishkin 2019] N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00049
GeoDesc-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 317.71
Rank: 105/130
0.486
Rank: 44/130
0.850
Rank: 42/130
0.48769
(±0.00064)
Rank: 83/130
394.67
Rank: 94/130
3863.42
Rank: 71/130
4.380
Rank: 64/130
0.440
Rank: 68/130
0.67682
(±0.00127)
Rank: 82/130
0.58225
Rank: 78/130
Challenge organizers (contact) sift8k geodesc-upright (128 float32: 512 bytes) Upright GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00057
HardNet-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 582.52
Rank: 36/130
0.486
Rank: 44/130
0.841
Rank: 57/130
0.53730
(±0.00082)
Rank: 44/130
480.13
Rank: 65/130
4167.34
Rank: 56/130
4.566
Rank: 31/130
0.416
Rank: 39/130
0.71542
(±0.00274)
Rank: 41/130
0.62636
Rank: 42/130
Challenge organizers (contact) sift8k hardnet-upright (128 float32: 512 bytes) Upright HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00047
GeoDesc w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 281.36
Rank: 110/130
0.472
Rank: 87/130
0.838
Rank: 64/130
0.47259
(±0.00033)
Rank: 88/130
350.43
Rank: 109/130
3601.67
Rank: 80/130
4.234
Rank: 92/130
0.447
Rank: 79/130
0.65553
(±0.00444)
Rank: 92/130
0.56406
Rank: 90/130
Challenge organizers (contact) sift8k geodesc (128 float32: 512 bytes) GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00558
SIFT + DeepOrientation + SOSNet ...
Size: 512 bytes. Matches: custom
2826.82 333.94
Rank: 101/130
0.284
Rank: 126/130
0.427
Rank: 124/130
0.01806
(±0.00000)
Rank: 128/130
341.75
Rank: 114/130
2003.56
Rank: 122/130
3.735
Rank: 119/130
0.593
Rank: 118/130
0.51048
(±0.00436)
Rank: 118/130
0.26427
Rank: 124/130
Fabio Bellavia (contact) sift deep-oriented-sosnet (128 float32: 512 bytes) SIFT (VLFeat implementation) [Lowe 2004] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 10) [Mishkin 2019] N/A N/A 20-05-11 is_submission, is_challenge_2020
Submission ID: 00079
LogPolar w/ DEGENSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 441.77
Rank: 66/130
0.472
Rank: 87/130
0.852
Rank: 34/130
0.53396
(±0.00086)
Rank: 46/130
415.77
Rank: 82/130
4054.60
Rank: 62/130
4.316
Rank: 80/130
0.432
Rank: 64/130
0.69284
(±0.00429)
Rank: 63/130
0.61340
Rank: 51/130
Challenge organizers (contact) sift8k logpolar96-fixed (128 float32: 512 bytes) LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00077
SOSNet-Upright w/ DEGENSAC (no F...
Size: 512 bytes. Matches: built-in
7829.63 508.38
Rank: 52/130
0.486
Rank: 44/130
0.877
Rank: 4/130
0.57385
(±0.00041)
Rank: 18/130
600.23
Rank: 36/130
4765.43
Rank: 27/130
4.584
Rank: 27/130
0.403
Rank: 27/130
0.73027
(±0.00284)
Rank: 25/130
0.65206
Rank: 24/130
Challenge organizers (contact) sift8k sosnet-upright (128 float32: 512 bytes) Upright SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-23 is_baseline
Submission ID: 00502
Guided matching with Upright roo...
Size: 512 bytes. Matches: custom
7829.27 491.96
Rank: 54/130
0.486
Rank: 29/130
0.828
Rank: 80/130
0.50310
(±0.01761)
Rank: 72/130
809.76
Rank: 20/130
5966.91
Rank: 13/130
4.462
Rank: 51/130
0.425
Rank: 55/130
0.71009
(±0.00414)
Rank: 45/130
0.60659
Rank: 64/130
(contact) sift upright-root-sift (128 float32: 512 bytes) In submission N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00015
Root-SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
7860.73 274.85
Rank: 112/130
0.472
Rank: 81/130
0.845
Rank: 50/130
0.48887
(±0.00011)
Rank: 82/130
437.74
Rank: 75/130
3814.80
Rank: 76/130
4.151
Rank: 104/130
0.458
Rank: 90/130
0.65067
(±0.00155)
Rank: 94/130
0.56977
Rank: 89/130
Challenge organizers (contact) sift-lowth rootsift (128 float32: 512 bytes) Root-SIFT with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00624
Guided-HardNet-OANet
Size: 512 bytes. Matches: custom
7829.63 765.34
Rank: 12/130
0.486
Rank: 44/130
0.820
Rank: 90/130
0.60261
(±0.00000)
Rank: 2/130
788.49
Rank: 21/130
6346.62
Rank: 6/130
4.682
Rank: 6/130
0.355
Rank: 1/130
0.78550
(±0.00157)
Rank: 1/130
0.69405
Rank: 2/130
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet-epoch2 (128 float32: 512 bytes) sift and hardnet with 8k features, first using the oanet trained from scratch then guided-matching, setting keypoint orientation to a constant value to increase performance. N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00542
sift and hardnet64 train scale(1...
Size: 512 bytes. Matches: built-in
7830.09 607.44
Rank: 28/130
0.486
Rank: 32/130
0.872
Rank: 12/130
0.58801
(±0.00037)
Rank: 6/130
950.74
Rank: 11/130
6286.36
Rank: 9/130
4.626
Rank: 17/130
0.390
Rank: 21/130
0.74662
(±0.00087)
Rank: 19/130
0.66732
Rank: 12/130
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-l2 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 138000), FLANN disabled N/A N/A 20-05-01 is_submission, is_challenge_2020
Submission ID: 00535
hardnetlib32P-Upright w/ DEGENSA...
Size: 512 bytes. Matches: built-in
7829.63 527.64
Rank: 46/130
0.486
Rank: 44/130
0.876
Rank: 6/130
0.57327
(±0.00078)
Rank: 20/130
506.73
Rank: 60/130
4230.98
Rank: 53/130
4.517
Rank: 45/130
0.418
Rank: 43/130
0.72171
(±0.00156)
Rank: 36/130
0.64749
Rank: 28/130
Chen Shen, Zhipeng Wang, Jun Zhang, Jingchao Zhou, Pengfei Xu (contact) sift8k-no-dups-mps hardnetlib32P (128 float32: 512 bytes) sift and hardnet with 8k features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-04-27 is_submission, is_challenge_2020
Submission ID: 00114
VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 348.95
Rank: 95/130
0.547
Rank: 12/130
0.800
Rank: 99/130
0.43347
(±0.00084)
Rank: 101/130
347.39
Rank: 111/130
3209.10
Rank: 103/130
4.126
Rank: 105/130
0.517
Rank: 112/130
0.58657
(±0.00120)
Rank: 113/130
0.51002
Rank: 109/130
Challenge organizers (contact) hessian vlsift (128 float32: 512 bytes) VL-Hess-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00563
HardNet64-train-all-SOS-812000
Size: 512 bytes. Matches: built-in
7830.09 612.51
Rank: 25/130
0.486
Rank: 32/130
0.869
Rank: 18/130
0.58664
(±0.00095)
Rank: 9/130
972.47
Rank: 8/130
6369.03
Rank: 4/130
4.612
Rank: 19/130
0.388
Rank: 18/130
0.74668
(±0.00251)
Rank: 18/130
0.66666
Rank: 13/130
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-sos-812000 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), sosnet64 with 128 descriptors(trained with sos loss and step 812000), FLANN disabled N/A N/A 20-05-13 is_submission, is_challenge_2020
Submission ID: 00607
HarrisZ (more kpt) + DeepOrienta...
Size: 512 bytes. Matches: custom
4478.94 746.02
Rank: 14/130
0.455
Rank: 112/130
0.699
Rank: 113/130
0.40725
(±0.00000)
Rank: 103/130
763.44
Rank: 24/130
2953.35
Rank: 113/130
4.414
Rank: 54/130
0.442
Rank: 69/130
0.69399
(±0.00179)
Rank: 62/130
0.55062
Rank: 93/130
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ (start scale 2) [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 0.75, degeneracy check true, confidence 0.98, max iter 500000) [Mishkin 2019] N/A N/A 20-06-03 is_submission, is_challenge_2020
Submission ID: 00600
ContextDesc Upright + Mutual Che...
Size: 512 bytes. Matches: custom
7830.09 624.55
Rank: 20/130
0.487
Rank: 25/130
0.847
Rank: 46/130
0.57344
(±0.00000)
Rank: 19/130
644.39
Rank: 30/130
5427.23
Rank: 19/130
4.700
Rank: 4/130
0.368
Rank: 5/130
0.77043
(±0.00133)
Rank: 4/130
0.67194
Rank: 6/130
Jiahui Zhang, Zixin Luo, Hongkai Chen (contact) contextdesc-upright contextdesc-upright (128 float32: 512 bytes) ContextDesc with 8000 SIFT features, using improved OANet matcher and DEGENSAC post-processing N/A N/A 20-05-29 is_submission, is_challenge_2020
Submission ID: 00055
HardNet-Upright w/ DEGENSAC (no ...
Size: 512 bytes. Matches: built-in
7829.63 527.59
Rank: 47/130
0.486
Rank: 44/130
0.876
Rank: 5/130
0.57279
(±0.00120)
Rank: 22/130
509.07
Rank: 56/130
4250.40
Rank: 51/130
4.548
Rank: 36/130
0.413
Rank: 37/130
0.72309
(±0.00141)
Rank: 32/130
0.64794
Rank: 27/130
Challenge organizers (contact) sift8k hardnet-upright (128 float32: 512 bytes) Upright HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00601
SIFT8k-HardNet64
Size: 512 bytes. Matches: built-in
6589.88 381.80
Rank: 82/130
0.467
Rank: 108/130
0.852
Rank: 35/130
0.52493
(±0.00044)
Rank: 53/130
420.36
Rank: 79/130
3586.82
Rank: 83/130
4.494
Rank: 48/130
0.429
Rank: 58/130
0.70580
(±0.00229)
Rank: 51/130
0.61536
Rank: 47/130
caoliang (contact) sift8k hardnet (128 float32: 512 bytes) SIFT up to 8000 keypoints, hardnet extract descriptors.Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-05-29 is_submission, is_challenge_2020
Submission ID: 00108
VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 171.63
Rank: 124/130
0.482
Rank: 75/130
0.848
Rank: 44/130
0.39839
(±0.00122)
Rank: 106/130
311.54
Rank: 123/130
3061.54
Rank: 107/130
4.105
Rank: 109/130
0.475
Rank: 103/130
0.62964
(±0.00428)
Rank: 99/130
0.51401
Rank: 107/130
Challenge organizers (contact) dogaffine vlsift (128 float32: 512 bytes) VL-DoGAff-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00594
CV-DoG-HardNet8-PT
Size: 512 bytes. Matches: built-in
7829.26 579.51
Rank: 37/130
0.484
Rank: 74/130
0.872
Rank: 13/130
0.58375
(±0.00046)
Rank: 11/130
573.22
Rank: 42/130
4489.23
Rank: 40/130
4.619
Rank: 18/130
0.403
Rank: 28/130
0.72432
(±0.00357)
Rank: 31/130
0.65403
Rank: 22/130
Milan Pultar, Dmytro Mishkin, Jiri Matas (contact) sift8k h8e512pt (128 float32: 512 bytes) HardNet8 with PCA compression N/A N/A 20-05-27 is_submission, is_challenge_2020
Submission ID: 00533
upright-sift8k-hardnet
Size: 512 bytes. Matches: built-in
6589.88 371.42
Rank: 86/130
0.467
Rank: 108/130
0.852
Rank: 39/130
0.52463
(±0.00070)
Rank: 54/130
404.02
Rank: 84/130
3501.62
Rank: 90/130
4.483
Rank: 49/130
0.428
Rank: 57/130
0.70123
(±0.00081)
Rank: 55/130
0.61293
Rank: 52/130
caoliang (contact) sift8k hardnet (128 float32: 512 bytes) SIFT up to 8000 keypoints, harnet extract descriptors.Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-24 is_submission, is_challenge_2020
Submission ID: 00614
ContextDesc Upright + Mutual Che...
Size: 512 bytes. Matches: custom
7830.09 647.63
Rank: 18/130
0.487
Rank: 25/130
0.846
Rank: 48/130
0.57826
(±0.00000)
Rank: 17/130
668.00
Rank: 28/130
5612.57
Rank: 17/130
4.666
Rank: 8/130
0.367
Rank: 4/130
0.77041
(±0.00298)
Rank: 5/130
0.67433
Rank: 5/130
Jiahui Zhang, Zixin Luo, Hongkai Chen (contact) contextdesc-upright contextdesc-upright (128 float32: 512 bytes) ContextDesc with 8000 SIFT features, using improved OANet matcher and DEGENSAC post-processing N/A N/A 20-05-31 is_submission, is_challenge_2020
Submission ID: 00053
HardNet w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 575.07
Rank: 39/130
0.472
Rank: 87/130
0.842
Rank: 56/130
0.55022
(±0.00001)
Rank: 34/130
426.84
Rank: 76/130
4001.40
Rank: 67/130
4.339
Rank: 75/130
0.419
Rank: 47/130
0.70962
(±0.00191)
Rank: 46/130
0.62992
Rank: 38/130
Challenge organizers (contact) sift8k hardnet (128 float32: 512 bytes) HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00005
SURF (OpenCV), DEGENSAC
Size: 256 bytes. Matches: built-in
7728.57 125.94
Rank: 128/130
0.432
Rank: 118/130
0.684
Rank: 114/130
0.24948
(±0.00065)
Rank: 116/130
928.29
Rank: 12/130
3455.28
Rank: 91/130
3.428
Rank: 123/130
0.733
Rank: 128/130
0.40490
(±0.00448)
Rank: 126/130
0.32719
Rank: 117/130
Challenge organizers (contact) surf-lowth surf (64 float32: 256 bytes) SURF with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00510
sift and hardnet64
Size: 512 bytes. Matches: built-in
7861.62 326.87
Rank: 102/130
0.472
Rank: 85/130
0.845
Rank: 51/130
0.51149
(±0.00018)
Rank: 67/130
521.58
Rank: 52/130
4587.88
Rank: 34/130
4.222
Rank: 95/130
0.424
Rank: 54/130
0.69458
(±0.00340)
Rank: 61/130
0.60304
Rank: 65/130
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-v2 (128 float32: 512 bytes) SIFT with 8000 keypoints, hardnet64 with 128 descriptors N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00559
SIFT + DeepOrientation + SOSNet ...
Size: 512 bytes. Matches: custom
2826.82 164.43
Rank: 125/130
0.284
Rank: 126/130
0.650
Rank: 116/130
0.05206
(±0.00000)
Rank: 125/130
167.59
Rank: 128/130
1549.91
Rank: 126/130
3.989
Rank: 118/130
0.622
Rank: 120/130
0.47700
(±0.00167)
Rank: 120/130
0.26453
Rank: 123/130
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) SIFT (VLFeat implementation) [Lowe 2004] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 5) [Mishkin 2019] N/A N/A 20-05-10 is_submission, is_challenge_2020
Submission ID: 00070
GeoDesc-Upright w/ MAGSAC (no FL...
Size: 512 bytes. Matches: built-in
7829.63 534.05
Rank: 44/130
0.486
Rank: 44/130
0.848
Rank: 45/130
0.52078
(±0.00028)
Rank: 59/130
458.60
Rank: 68/130
4146.83
Rank: 59/130
4.412
Rank: 55/130
0.432
Rank: 63/130
0.70442
(±0.00189)
Rank: 53/130
0.61260
Rank: 54/130
Challenge organizers (contact) sift8k geodesc-upright (128 float32: 512 bytes) Upright GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00045
D2-Net (multi-scale), DEGENSAC
Size: 2048 bytes. Matches: built-in
6924.14 338.82
Rank: 100/130
0.321
Rank: 124/130
0.386
Rank: 125/130
0.20486
(±0.00062)
Rank: 119/130
740.77
Rank: 25/130
5696.95
Rank: 15/130
3.326
Rank: 127/130
0.660
Rank: 123/130
0.42782
(±0.00544)
Rank: 123/130
0.31634
Rank: 118/130
Challenge organizers (contact) d2net-multiscale d2net-multiscale (512 float32: 2048 bytes) D2-Net, multi-scale model, up to 8000 features. Trained on the MegaDepth dataset, removing scenes which overlap with the Phototourism test set. Stereo with DEGENSAC and optimal parameters http://openaccess.thecvf.com/content_CVPR_2019/papers/Dusmanu_D2-Net_A_Trainable_CNN_for_Joint_Description_and_Detection_of_CVPR_2019_paper.pdf https://github.com/mihaidusmanu/d2-net 20-04-22 is_baseline
Submission ID: 00052
HardNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 363.51
Rank: 89/130
0.472
Rank: 87/130
0.853
Rank: 33/130
0.52325
(±0.00098)
Rank: 56/130
402.15
Rank: 85/130
3895.00
Rank: 69/130
4.345
Rank: 70/130
0.426
Rank: 56/130
0.69653
(±0.00108)
Rank: 59/130
0.60989
Rank: 58/130
Challenge organizers (contact) sift8k hardnet (128 float32: 512 bytes) HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00549
SIFT + DeepOrientation + SOSNet ...
Size: 512 bytes. Matches: custom
2826.82 216.23
Rank: 120/130
0.284
Rank: 126/130
0.456
Rank: 123/130
0.14544
(±0.00000)
Rank: 122/130
221.40
Rank: 127/130
1839.87
Rank: 123/130
3.608
Rank: 120/130
0.641
Rank: 121/130
0.47073
(±0.00158)
Rank: 121/130
0.30808
Rank: 120/130
Fabio Bellavia (contact) sift deep-oriented-sosnet (128 float32: 512 bytes) SIFT (VLFeat implementation) [Lowe 2004] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 0.75, degeneracy check true) [Mishkin 2019] N/A N/A 20-05-03 is_submission, is_challenge_2020
Submission ID: 00113
VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 290.24
Rank: 109/130
0.547
Rank: 12/130
0.814
Rank: 95/130
0.44501
(±0.00077)
Rank: 100/130
347.39
Rank: 111/130
3209.10
Rank: 103/130
4.126
Rank: 105/130
0.519
Rank: 115/130
0.58657
(±0.00120)
Rank: 113/130
0.51579
Rank: 105/130
Challenge organizers (contact) hessian vlsift (128 float32: 512 bytes) VL-Hess-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00567
Guided-HardNet32-v1-lib-qht-p
Size: 512 bytes. Matches: custom
7829.63 520.40
Rank: 48/130
0.486
Rank: 44/130
0.875
Rank: 8/130
0.58509
(±0.00000)
Rank: 10/130
536.93
Rank: 47/130
4685.16
Rank: 32/130
4.645
Rank: 12/130
0.376
Rank: 9/130
0.75702
(±0.00159)
Rank: 12/130
0.67105
Rank: 7/130
Chen Shen, Zhipeng Wang, Jun Zhang, Zhiwei Ruan, Zhongkun Chen, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet32-v1-lib-qht-p (128 float32: 512 bytes) sift and sosnet with 8k features, using the guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-15 is_submission, is_challenge_2020
Submission ID: 00547
SIFT + DeepOrientation + SOSNet ...
Size: 512 bytes. Matches: custom
2826.82 1078.85
Rank: 4/130
0.284
Rank: 126/130
0.166
Rank: 129/130
0.04936
(±0.00000)
Rank: 126/130
1106.68
Rank: 5/130
2494.00
Rank: 119/130
3.400
Rank: 124/130
0.646
Rank: 122/130
0.43054
(±0.00185)
Rank: 122/130
0.23995
Rank: 126/130
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) SIFT (VLFeat implementation) [Lowe 2004] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 15, degeneracy check true) [Mishkin 2019] N/A N/A 20-05-03 is_submission, is_challenge_2020
Submission ID: 00537
SIFT + DeepOrientation + SOSNet ...
Size: 512 bytes. Matches: custom
2826.82 1476.71
Rank: 1/130
0.284
Rank: 126/130
0.145
Rank: 130/130
0.00178
(±0.00000)
Rank: 130/130
1506.82
Rank: 2/130
2538.44
Rank: 118/130
3.388
Rank: 126/130
0.669
Rank: 124/130
0.42682
(±0.00259)
Rank: 124/130
0.21430
Rank: 127/130
Fabio Bellavia (contact) sift deep-oriented-sosnet (128 float32: 512 bytes) SIFT (VLFeat implementation) [Lowe 2004] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] N/A N/A 20-05-01 is_submission, is_challenge_2020
Submission ID: 00110
VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 317.13
Rank: 106/130
0.482
Rank: 75/130
0.820
Rank: 91/130
0.46657
(±0.00079)
Rank: 92/130
311.54
Rank: 123/130
3061.54
Rank: 107/130
4.105
Rank: 109/130
0.475
Rank: 102/130
0.62964
(±0.00428)
Rank: 99/130
0.54810
Rank: 95/130
Challenge organizers (contact) dogaffine vlsift (128 float32: 512 bytes) VL-DoGAff-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00072
L2-Net-Upright w/ DEGENSAC (no F...
Size: 512 bytes. Matches: built-in
7829.63 435.68
Rank: 68/130
0.486
Rank: 44/130
0.859
Rank: 28/130
0.54497
(±0.00003)
Rank: 36/130
395.53
Rank: 90/130
3603.85
Rank: 78/130
4.382
Rank: 61/130
0.452
Rank: 86/130
0.68491
(±0.00338)
Rank: 74/130
0.61494
Rank: 49/130
Challenge organizers (contact) sift8k l2net-upright (128 float32: 512 bytes) Upright L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-22 is_baseline
Submission ID: 00590
Guided-HardNet-epoch4
Size: 512 bytes. Matches: custom
7829.63 586.24
Rank: 34/130
0.486
Rank: 44/130
0.875
Rank: 7/130
0.59919
(±0.00000)
Rank: 3/130
604.81
Rank: 35/130
5062.27
Rank: 23/130
4.710
Rank: 3/130
0.369
Rank: 7/130
0.76219
(±0.00253)
Rank: 7/130
0.68069
Rank: 3/130
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet-epoch4 (128 float32: 512 bytes) sift and hardnet with 8k features, using the modified guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-24 is_submission, is_challenge_2020
Submission ID: 00117
VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 299.02
Rank: 107/130
0.577
Rank: 6/130
0.831
Rank: 72/130
0.46793
(±0.00035)
Rank: 91/130
350.69
Rank: 106/130
3327.71
Rank: 98/130
4.076
Rank: 112/130
0.487
Rank: 105/130
0.60691
(±0.00338)
Rank: 107/130
0.53742
Rank: 100/130
Challenge organizers (contact) hessian affnetvlsift (128 float32: 512 bytes) VL-HessAffNet-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00610
Hardnet-Upright-AdaLAM
Size: 512 bytes. Matches: custom
6556.61 627.71
Rank: 19/130
0.442
Rank: 115/130
0.828
Rank: 79/130
0.58300
(±0.00000)
Rank: 12/130
645.47
Rank: 29/130
5074.91
Rank: 22/130
4.575
Rank: 30/130
0.361
Rank: 3/130
0.77056
(±0.00064)
Rank: 3/130
0.67678
Rank: 4/130
Luca Cavalli, Viktor Larsson, Martin Oswald, Torsten Sattler, Marc Pollefeys (contact) sift-def hardnet-upright (128 float32: 512 bytes) Using upright Hardnet descriptors with 8000 features, nearest neighbor matching and outlier rejection enforcing local affine consistency within a confidence-based adaptive error tolerance. Matches post-processed with DEGENSAC. https://arxiv.org/abs/2006.04250 N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00074
L2-Net w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 481.03
Rank: 55/130
0.472
Rank: 87/130
0.830
Rank: 78/130
0.52519
(±0.00039)
Rank: 51/130
339.00
Rank: 115/130
3424.94
Rank: 92/130
4.206
Rank: 97/130
0.448
Rank: 82/130
0.66445
(±0.00213)
Rank: 87/130
0.59482
Rank: 74/130
Challenge organizers (contact) sift8k l2net (128 float32: 512 bytes) L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-23 is_baseline
Submission ID: 00595
SIFT8k_8000_HardNet64-train-all-...
Size: 512 bytes. Matches: built-in
7830.09 600.42
Rank: 31/130
0.486
Rank: 32/130
0.867
Rank: 20/130
0.58265
(±0.00102)
Rank: 14/130
786.40
Rank: 22/130
5595.19
Rank: 18/130
4.603
Rank: 22/130
0.390
Rank: 20/130
0.74372
(±0.00317)
Rank: 23/130
0.66319
Rank: 16/130
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-ben86-l2-210000 (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 210000 and iou 0.86), FLANN disabled N/A N/A 20-05-27 is_submission, is_challenge_2020
Submission ID: 00618
Sift-Sem + HyNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7530.17 476.84
Rank: 57/130
0.490
Rank: 19/130
0.882
Rank: 1/130
0.57281
(±0.00047)
Rank: 21/130
809.82
Rank: 19/130
5868.33
Rank: 14/130
4.585
Rank: 26/130
0.390
Rank: 22/130
0.74557
(±0.00393)
Rank: 21/130
0.65919
Rank: 19/130
Barroso-Laguna, Axel and Tian, Yurun, Ng, Tony and Mikolajczyk, Krystian (contact) sift-semantics hynet (128 float32: 512 bytes) N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00521
SIFT8k-gift
Size: 512 bytes. Matches: built-in
6589.48 426.77
Rank: 71/130
0.467
Rank: 111/130
0.830
Rank: 74/130
0.48128
(±0.00056)
Rank: 86/130
452.18
Rank: 70/130
3770.17
Rank: 77/130
4.595
Rank: 25/130
0.444
Rank: 76/130
0.67443
(±0.00253)
Rank: 84/130
0.57785
Rank: 80/130
Chen Shen (contact) sift-def gift (128 float32: 512 bytes) sift-nodups, gift with 8k features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-04-26 is_submission, is_challenge_2020
Submission ID: 00566
Guided-HardNet2qhtp.json
Size: 512 bytes. Matches: custom
7829.63 517.72
Rank: 49/130
0.486
Rank: 44/130
0.874
Rank: 9/130
0.58251
(±0.00000)
Rank: 15/130
448.12
Rank: 72/130
4072.82
Rank: 61/130
4.607
Rank: 20/130
0.387
Rank: 17/130
0.74640
(±0.00246)
Rank: 20/130
0.66445
Rank: 14/130
Chen Shen, Zhipeng Wang, Jun Zhang, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnetnd32qhtp (128 float32: 512 bytes) sift and hardnet with 8k features, using the guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-15 is_submission, is_challenge_2020
Submission ID: 00572
Guided HardNet qht
Size: 512 bytes. Matches: custom
7829.63 885.09
Rank: 8/130
0.486
Rank: 44/130
0.774
Rank: 106/130
0.54663
(±0.00000)
Rank: 35/130
538.28
Rank: 45/130
4687.01
Rank: 31/130
4.644
Rank: 13/130
0.376
Rank: 10/130
0.75721
(±0.00140)
Rank: 11/130
0.65192
Rank: 25/130
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift8k hardnet-qht (128 float32: 512 bytes) sift and hardnet with 8k features, using the guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-20 is_submission, is_challenge_2020
Submission ID: 00541
sift and sosnet64 train scale(12...
Size: 512 bytes. Matches: built-in
7830.09 603.56
Rank: 30/130
0.486
Rank: 32/130
0.866
Rank: 21/130
0.58268
(±0.00061)
Rank: 13/130
957.95
Rank: 10/130
6312.38
Rank: 8/130
4.607
Rank: 21/130
0.385
Rank: 14/130
0.74476
(±0.00137)
Rank: 22/130
0.66372
Rank: 15/130
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet64-train-all-sos (128 float32: 512 bytes) SIFT with 8000 keypoints(scale 12), sosnet64 with 128 descriptors(trained with sos loss and step 348000), FLANN disabled N/A N/A 20-05-01 is_submission, is_challenge_2020
Submission ID: 00102
KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 815.40
Rank: 10/130
0.582
Rank: 2/130
0.788
Rank: 101/130
0.47395
(±0.00073)
Rank: 87/130
356.21
Rank: 102/130
3366.01
Rank: 95/130
4.319
Rank: 77/130
0.464
Rank: 96/130
0.64829
(±0.00174)
Rank: 95/130
0.56112
Rank: 92/130
Challenge organizers (contact) keynettuned vlhardnet (128 float32: 512 bytes) KeyNet-HardNet with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with MAGSAC N/A N/A 20-05-04 is_baseline
Submission ID: 00556
HarrisZ + DeepOrientation + SOSN...
Size: 512 bytes. Matches: custom
2410.18 385.16
Rank: 81/130
0.368
Rank: 119/130
0.646
Rank: 118/130
0.04192
(±0.00000)
Rank: 127/130
392.52
Rank: 96/130
1475.19
Rank: 127/130
4.454
Rank: 52/130
0.601
Rank: 119/130
0.48498
(±0.00118)
Rank: 119/130
0.26345
Rank: 125/130
Fabio Bellavia (contact) hz deep-oriented-sosnet (128 float32: 512 bytes) HarrisZ [Bellavia et al. 2011] + Deep patch orientation [Yi et al. 2015] + SOSNet [Tian at al. 2019] (weights: sosnet-32x32-hpatches_a.pth) + Blob Matching [unpublished] + Delaunay Triangulation Matching (DTM) [unpublished] + PyRANSAC (threshold 5) [Mishkin 2019] N/A N/A 20-05-11 is_submission, is_challenge_2020
Submission ID: 00073
L2-Net-Upright w/ MAGSAC (no FLA...
Size: 512 bytes. Matches: built-in
7829.63 577.91
Rank: 38/130
0.486
Rank: 44/130
0.837
Rank: 67/130
0.53911
(±0.00062)
Rank: 41/130
395.53
Rank: 90/130
3603.85
Rank: 78/130
4.382
Rank: 61/130
0.455
Rank: 88/130
0.68491
(±0.00338)
Rank: 74/130
0.61201
Rank: 55/130
Challenge organizers (contact) sift8k l2net-upright (128 float32: 512 bytes) Upright L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-22 is_baseline
Submission ID: 00080
LogPolar w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 376.42
Rank: 84/130
0.472
Rank: 87/130
0.837
Rank: 65/130
0.50188
(±0.00044)
Rank: 73/130
399.70
Rank: 87/130
4036.37
Rank: 64/130
4.340
Rank: 73/130
0.429
Rank: 59/130
0.69100
(±0.00154)
Rank: 66/130
0.59644
Rank: 72/130
Challenge organizers (contact) sift8k logpolar96-fixed (128 float32: 512 bytes) LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00006
Upright SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
7829.24 319.29
Rank: 104/130
0.487
Rank: 23/130
0.830
Rank: 76/130
0.48742
(±0.00042)
Rank: 84/130
525.57
Rank: 49/130
4147.20
Rank: 58/130
4.258
Rank: 90/130
0.460
Rank: 92/130
0.65664
(±0.00231)
Rank: 91/130
0.57203
Rank: 87/130
Challenge organizers (contact) sift-lowth sift-upright (128 float32: 512 bytes) Upright SIFT with (up to) 8000 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00059
L2-Net w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 297.34
Rank: 108/130
0.472
Rank: 87/130
0.835
Rank: 70/130
0.49523
(±0.00042)
Rank: 77/130
314.54
Rank: 121/130
3314.19
Rank: 101/130
4.204
Rank: 99/130
0.467
Rank: 99/130
0.65665
(±0.00287)
Rank: 89/130
0.57594
Rank: 83/130
Challenge organizers (contact) sift8k l2net (128 float32: 512 bytes) L2-Net descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. http://openaccess.thecvf.com/content_cvpr_2017/papers/Tian_L2-Net_Deep_Learning_CVPR_2017_paper.pdf https://github.com/yuruntian/L2-Net 20-04-22 is_baseline
Submission ID: 00050
GeoDesc-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 410.23
Rank: 74/130
0.486
Rank: 44/130
0.827
Rank: 81/130
0.48259
(±0.00060)
Rank: 85/130
394.67
Rank: 94/130
3863.42
Rank: 71/130
4.380
Rank: 64/130
0.443
Rank: 73/130
0.67682
(±0.00127)
Rank: 82/130
0.57971
Rank: 79/130
Challenge organizers (contact) sift8k geodesc-upright (128 float32: 512 bytes) Upright GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00054
HardNet w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 477.64
Rank: 56/130
0.472
Rank: 87/130
0.830
Rank: 77/130
0.51809
(±0.00053)
Rank: 62/130
402.15
Rank: 85/130
3895.00
Rank: 69/130
4.345
Rank: 70/130
0.423
Rank: 53/130
0.69653
(±0.00108)
Rank: 59/130
0.60731
Rank: 63/130
Challenge organizers (contact) sift8k hardnet (128 float32: 512 bytes) HardNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00067
GeoDesc w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 453.39
Rank: 63/130
0.472
Rank: 87/130
0.835
Rank: 68/130
0.50559
(±0.00081)
Rank: 71/130
395.12
Rank: 92/130
3838.97
Rank: 73/130
4.264
Rank: 88/130
0.442
Rank: 71/130
0.68032
(±0.00099)
Rank: 80/130
0.59295
Rank: 75/130
Challenge organizers (contact) sift8k geodesc (128 float32: 512 bytes) GeoDesc descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. FLANN disabled. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00540
ASLV2+OANetV2+DEGENSAC
Size: 512 bytes. Matches: custom
5982.87 1044.40
Rank: 5/130
0.550
Rank: 11/130
0.758
Rank: 108/130
0.50173
(±0.00000)
Rank: 74/130
1067.11
Rank: 6/130
4587.47
Rank: 35/130
4.948
Rank: 2/130
0.386
Rank: 16/130
0.75275
(±0.00186)
Rank: 13/130
0.62724
Rank: 41/130
Jiahui Zhang, Zixin Luo, Hongkai Chen (contact) aslv2 aslv2 (128 float32: 512 bytes) ASL detector and descriptor, 8k keypoints, using improved OANet matcher and DEGENSAC post-processing N/A N/A 20-05-01 is_submission, is_challenge_2020
Submission ID: 00514
UprightRootSIFT-AdaLAM
Size: 512 bytes. Matches: custom
6449.42 619.87
Rank: 23/130
0.436
Rank: 116/130
0.758
Rank: 109/130
0.31372
(±0.00000)
Rank: 113/130
640.08
Rank: 31/130
5286.27
Rank: 21/130
4.343
Rank: 72/130
0.398
Rank: 25/130
0.72130
(±0.00085)
Rank: 37/130
0.51751
Rank: 104/130
Luca Cavalli, Viktor Larsson, Martin Oswald, Torsten Sattler, Marc Pollefeys (contact) sift-def rootsift-upright (128 float32: 512 bytes) Using upright RootSIFT with 8000 features, nearest neighbor matching and outlier rejection enforcing local affine consistency within a confidence-based adaptive error tolerance. https://arxiv.org/abs/2006.04250 N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00511
r2d2-wasfi
Size: 512 bytes. Matches: built-in
7861.03 276.55
Rank: 111/130
0.503
Rank: 17/130
0.762
Rank: 107/130
0.37893
(±0.00023)
Rank: 109/130
416.35
Rank: 81/130
2834.69
Rank: 114/130
4.293
Rank: 84/130
0.480
Rank: 104/130
0.64359
(±0.00103)
Rank: 98/130
0.51126
Rank: 108/130
Chen Shen (contact) r2d2 r2d2-wasfi-epoch25-pretrained-n16-8k (128 float32: 512 bytes) r2d2 with 8k features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00083
LogPolar-Upright w/ DEGENSAC (no...
Size: 512 bytes. Matches: built-in
7829.63 543.18
Rank: 42/130
0.486
Rank: 44/130
0.865
Rank: 24/130
0.55102
(±0.00015)
Rank: 33/130
505.37
Rank: 61/130
4414.11
Rank: 45/130
4.518
Rank: 43/130
0.422
Rank: 50/130
0.71092
(±0.00251)
Rank: 43/130
0.63097
Rank: 37/130
Challenge organizers (contact) sift8k logpolar96-fixed-upright (128 float32: 512 bytes) Upright LogPolar descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00064
SOSNet w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 444.55
Rank: 65/130
0.472
Rank: 87/130
0.831
Rank: 73/130
0.51402
(±0.00057)
Rank: 65/130
440.21
Rank: 73/130
4178.82
Rank: 54/130
4.378
Rank: 66/130
0.422
Rank: 51/130
0.70078
(±0.00375)
Rank: 56/130
0.60740
Rank: 62/130
Challenge organizers (contact) sift8k sosnet (128 float32: 512 bytes) SOSNet descriptors on DoG features (OpenCV), extracted with a low detection threshold to generate up to 8000 points. Using the built-in matcher: bidirectional filter with the both strategy. MAGSAC for stereo. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00602
retrained sosnet
Size: 512 bytes. Matches: built-in
7861.11 340.97
Rank: 98/130
0.472
Rank: 87/130
0.840
Rank: 60/130
0.48990
(±0.00046)
Rank: 81/130
623.55
Rank: 33/130
4812.36
Rank: 26/130
4.233
Rank: 94/130
0.458
Rank: 91/130
0.66502
(±0.00158)
Rank: 86/130
0.57746
Rank: 82/130
sosnet (contact) siftdef sosnet (128 float32: 512 bytes) default setting and retrained sosnet N/A N/A 20-05-30 is_submission, is_challenge_2020
Submission ID: 00104
VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 179.68
Rank: 123/130
0.490
Rank: 20/130
0.835
Rank: 69/130
0.39987
(±0.00068)
Rank: 104/130
324.62
Rank: 118/130
3030.67
Rank: 110/130
4.173
Rank: 101/130
0.467
Rank: 100/130
0.62829
(±0.00112)
Rank: 103/130
0.51408
Rank: 106/130
Challenge organizers (contact) dog vlsift (128 float32: 512 bytes) VL-DoG-SIFT with 8000 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-04 is_baseline

Phototourism: restricted keypoints, large descriptors (2048 bytes)

Note: entries with the same multi-view configuration may seem duplicated. This is normal: performance is averaged across tasks.
Stereo Multiview Avg.
Method NF NI Rep.
(3 pix.)
MS
(3 pix.)
mAA
(at 10o)
NM NL TL ATE mAA
(at 100)
mAA
(at 100)
Authors Keypoint Descriptor Summary Paper Website Processing date Flags
Submission ID: 00555
SuperPoint+GIFT+Graph Motion Coh...
Size: 512 bytes. Matches: custom
1940.85 281.06
Rank: 9/86
0.356
Rank: 26/86
0.676
Rank: 68/86
0.42283
(±0.00000)
Rank: 33/86
290.06
Rank: 21/86
1887.70
Rank: 17/86
4.567
Rank: 15/86
0.420
Rank: 8/86
0.70395
(±0.00239)
Rank: 7/86
0.56339
Rank: 16/86
Yuan Liu (contact) superpoint-2k scale-gift (128 float32: 512 bytes) Detecting by SuperPoint, descriptors from GIFT, matching by Graph Motion Coherence Network, geometry estimated by DEGENSAC with inlier threshold 1.0 N/A N/A 20-05-10 is_submission, is_challenge_2020
Submission ID: 00038
SOSNet, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 134.59
Rank: 43/86
0.323
Rank: 62/86
0.760
Rank: 57/86
0.38654
(±0.00056)
Rank: 49/86
145.47
Rank: 75/86
1271.41
Rank: 58/86
4.045
Rank: 61/86
0.526
Rank: 56/86
0.56072
(±0.00077)
Rank: 56/86
0.47363
Rank: 52/86
Challenge organizers (contact) sift-def sosnet (128 float32: 512 bytes) SOSNet descriptors extracted on SIFT keypoints and MAGSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00705
Upright SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.72 104.74
Rank: 67/86
0.333
Rank: 38/86
0.764
Rank: 55/86
0.37292
(±0.00081)
Rank: 56/86
204.13
Rank: 39/86
1293.75
Rank: 55/86
4.008
Rank: 66/86
0.567
Rank: 70/86
0.52808
(±0.00416)
Rank: 65/86
0.45050
Rank: 61/86
Challenge organizers (contact) sift-def sift-upright (128 float32: 512 bytes) Upright SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-06-01 is_baseline
Submission ID: 00623
HardNet-epoch2-OANet
Size: 512 bytes. Matches: custom
2047.76 247.45
Rank: 12/86
0.347
Rank: 28/86
0.765
Rank: 53/86
0.44717
(±0.00000)
Rank: 21/86
254.11
Rank: 29/86
1898.45
Rank: 16/86
4.503
Rank: 18/86
0.437
Rank: 14/86
0.67958
(±0.00183)
Rank: 14/86
0.56337
Rank: 17/86
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift2k hardnet-epoch2 (128 float32: 512 bytes) sift and hardnet with 2k features, using the oanet trained from scratch and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00007
L2-Net, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 87.04
Rank: 77/86
0.323
Rank: 68/86
0.779
Rank: 40/86
0.36265
(±0.00090)
Rank: 58/86
138.97
Rank: 77/86
1211.49
Rank: 64/86
3.948
Rank: 70/86
0.537
Rank: 61/86
0.52019
(±0.00120)
Rank: 67/86
0.44142
Rank: 63/86
Challenge organizers (contact) sift-def l2net (128 float32: 512 bytes) L2-Net descriptors extracted on SIFT keypoints and DEGENSAC for stereo. Please refer to the baselines repository (linked) for details. http://www.nlpr.ia.ac.cn/fanbin/pub/L2-Net_CVPR17.pdf https://github.com/vcg-uvic/image-matching-benchmark-baselines 20-04-23 is_baseline
Submission ID: 00606
HardNet-epoch2-OANet
Size: 512 bytes. Matches: custom
2047.76 244.62
Rank: 15/86
0.347
Rank: 28/86
0.773
Rank: 47/86
0.46389
(±0.00000)
Rank: 12/86
251.11
Rank: 31/86
1916.44
Rank: 15/86
4.501
Rank: 19/86
0.428
Rank: 12/86
0.69407
(±0.00216)
Rank: 11/86
0.57898
Rank: 9/86
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift2k hardnet-epoch2 (128 float32: 512 bytes) sift and hardnet with 2k features, using the oanet and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-30 is_submission, is_challenge_2020
Submission ID: 00604
Key.Net+GIFT+GMCNet+DEGENSAC
Size: 512 bytes. Matches: custom
2046.07 377.20
Rank: 4/86
0.450
Rank: 4/86
0.776
Rank: 43/86
0.45282
(±0.00000)
Rank: 17/86
386.63
Rank: 14/86
1474.77
Rank: 32/86
4.663
Rank: 8/86
0.420
Rank: 7/86
0.70500
(±0.00294)
Rank: 6/86
0.57891
Rank: 10/86
Yuan Liu (contact) keynet-2k scale-gift (128 float32: 512 bytes) Detecting by Key.Net, descriptors from GIFT, matching by Graph Motion Coherence Network, geometry estimated by DEGENSAC with inlier threshold 0.7 N/A N/A 20-05-30 is_submission, is_challenge_2020
Submission ID: 00703
Upright Root-SIFT (OpenCV), DEGE...
Size: 512 bytes. Matches: built-in
1892.72 112.27
Rank: 60/86
0.333
Rank: 38/86
0.782
Rank: 37/86
0.39860
(±0.00077)
Rank: 43/86
199.34
Rank: 42/86
1341.66
Rank: 46/86
4.090
Rank: 57/86
0.518
Rank: 51/86
0.56230
(±0.00234)
Rank: 54/86
0.48045
Rank: 47/86
Challenge organizers (contact) sift-def rootsift-upright (128 float32: 512 bytes) Upright Root-SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-06-01 is_baseline
Submission ID: 00516
Key.Net (MS 4-1, Pyramid) + X-Ne...
Size: 512 bytes. Matches: built-in
2039.11 265.52
Rank: 11/86
0.447
Rank: 9/86
0.776
Rank: 45/86
0.40559
(±0.00112)
Rank: 38/86
336.81
Rank: 16/86
1559.74
Rank: 27/86
4.536
Rank: 16/86
0.461
Rank: 25/86
0.66406
(±0.00521)
Rank: 20/86
0.53483
Rank: 29/86
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) keynet-41-pyramidv2 x-net-lib (128 float32: 512 bytes) N/A N/A 20-04-24 is_submission, is_challenge_2020
Submission ID: 00002
SURF (OpenCV), DEGENSAC
Size: 256 bytes. Matches: built-in
2010.88 71.90
Rank: 80/86
0.321
Rank: 74/86
0.634
Rank: 69/86
0.20761
(±0.00029)
Rank: 75/86
300.02
Rank: 20/86
1208.06
Rank: 67/86
3.549
Rank: 75/86
0.695
Rank: 75/86
0.37845
(±0.00399)
Rank: 75/86
0.29303
Rank: 75/86
Challenge organizers (contact) surf-def surf (64 float32: 256 bytes) SURF with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00544
Key.Net + X-Net(Lib) w/ DEGENSAC
Size: 512 bytes. Matches: built-in
2039.11 194.10
Rank: 17/86
0.447
Rank: 9/86
0.814
Rank: 12/86
0.42332
(±0.00096)
Rank: 32/86
312.61
Rank: 18/86
1515.67
Rank: 29/86
4.567
Rank: 14/86
0.448
Rank: 17/86
0.66559
(±0.00233)
Rank: 19/86
0.54445
Rank: 22/86
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) keynet x-net-lib (128 float32: 512 bytes) N/A N/A 20-04-28 is_submission, is_challenge_2020
Submission ID: 00523
SEKD
Size: 512 bytes. Matches: built-in
2043.44 116.81
Rank: 54/86
0.386
Rank: 17/86
0.799
Rank: 26/86
0.42811
(±0.00096)
Rank: 30/86
166.41
Rank: 59/86
1163.31
Rank: 71/86
4.427
Rank: 23/86
0.456
Rank: 22/86
0.64919
(±0.00171)
Rank: 23/86
0.53865
Rank: 26/86
Yafei Song, Ling Cai, Mingyang Li (contact) sekd sekd (128 float32: 512 bytes) Our method named SEKD: Self-Evolving Keypoint Detection and Description, where the SEKD model is trained using COCO test images. In this submission each image has up to 2048 SEKD keypoints, and 128-dim float descriptor. We use the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC. N/A N/A 20-04-29 is_submission, is_challenge_2020
Submission ID: 00575
SIFT2k-2048-HardNet64-rain-all-s...
Size: 512 bytes. Matches: built-in
2047.77 161.77
Rank: 25/86
0.339
Rank: 34/86
0.814
Rank: 10/86
0.43543
(±0.00051)
Rank: 26/86
255.45
Rank: 27/86
1708.29
Rank: 22/86
4.327
Rank: 34/86
0.451
Rank: 19/86
0.64653
(±0.00221)
Rank: 24/86
0.54098
Rank: 24/86
Ximin Zheng, Sheng He, Hualong Shi (contact) sift2k sift2k-2048-hardnet64-train-all-sos-812000 (128 float32: 512 bytes) SIFT with 2048 keypoints(scale 12), sosnet64 with 128 descriptors(trained with sos loss and step 812000), FLANN disabled N/A N/A 20-05-20 is_submission, is_challenge_2020
Submission ID: 00041
SuperPoint (2k features, NMS=3),...
Size: 1024 bytes. Matches: built-in
2048.00 117.13
Rank: 51/86
0.368
Rank: 21/86
0.628
Rank: 73/86
0.28679
(±0.00056)
Rank: 70/86
156.04
Rank: 66/86
1121.20
Rank: 75/86
4.310
Rank: 40/86
0.566
Rank: 69/86
0.53829
(±0.00021)
Rank: 63/86
0.41254
Rank: 68/86
Challenge organizers (contact) superpoint-nms3-r1200 superpoint-nms3-r1200 (256 float32: 1024 bytes) SuperPoint (initial release based on Pytorch). Lowered detection threshold to obtain up to 2k features per image. Using the built-in Non-Maxima Suppression filter at 3 pixels. Images are resized to 1200 pixels on the largest side. https://arxiv.org/abs/1712.07629 https://github.com/MagicLeapResearch/SuperPointPretrainedNetwork 20-04-23 is_baseline
Submission ID: 00599
SEKD
Size: 512 bytes. Matches: built-in
2043.44 129.46
Rank: 47/86
0.386
Rank: 17/86
0.809
Rank: 16/86
0.45066
(±0.00057)
Rank: 18/86
176.60
Rank: 51/86
1209.63
Rank: 66/86
4.437
Rank: 22/86
0.454
Rank: 20/86
0.66094
(±0.00227)
Rank: 21/86
0.55580
Rank: 18/86
Yafei Song, Ling Cai, Jia Li, Yonghong Tian, Mingyang Li (contact) sekd sekd (128 float32: 512 bytes) Our method named SEKD: Self-Evolving Keypoint Detection and Description, where the SEKD model is trained using COCO validation set. In this submission each image has up to 2048 SEKD keypoints, and 128-dim float descriptor. We use the built-in matcher (bidirectional filter with the both strategy, without flann, optimal inlier and ratio test thresholds) with DEGENSAC. N/A N/A 20-05-29 is_submission, is_challenge_2020
Submission ID: 00001
HardNet, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 109.52
Rank: 63/86
0.323
Rank: 62/86
0.776
Rank: 44/86
0.38578
(±0.00131)
Rank: 51/86
153.29
Rank: 69/86
1306.64
Rank: 52/86
4.026
Rank: 63/86
0.539
Rank: 64/86
0.55573
(±0.00175)
Rank: 60/86
0.47075
Rank: 54/86
Challenge organizers (contact) sift-def hardnet (128 float32: 512 bytes) HardNet descriptors extracted on SIFT keypoints and DEGENSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00565
SIFT2k_2000_HardNet64-train-all-...
Size: 512 bytes. Matches: built-in
1999.83 157.70
Rank: 27/86
0.336
Rank: 36/86
0.813
Rank: 13/86
0.43188
(±0.00005)
Rank: 27/86
248.91
Rank: 32/86
1666.27
Rank: 24/86
4.323
Rank: 35/86
0.465
Rank: 28/86
0.64404
(±0.00271)
Rank: 26/86
0.53796
Rank: 27/86
Ximin Zheng, Sheng He, Hualong Shi (contact) sift2k sift2k-2000-hardnet64-train-all-sos-812000 (128 float32: 512 bytes) SIFT with 2000 keypoints(scale 12), sosnet64 with 128 descriptors(trained with sos loss and step 812000), FLANN disabled N/A N/A 20-05-13 is_submission, is_challenge_2020
Submission ID: 00620
pffNet + SuperPoint + MAGSAC
Size: 512 bytes. Matches: built-in
1267.22 57.48
Rank: 84/86
0.341
Rank: 32/86
0.632
Rank: 70/86
0.22549
(±0.00077)
Rank: 73/86
131.91
Rank: 79/86
893.88
Rank: 83/86
4.339
Rank: 29/86
0.531
Rank: 59/86
0.55935
(±0.00143)
Rank: 58/86
0.39242
Rank: 72/86
Jongmin Lee, Seungwook Kim, Yoonwoo Jeong (contact) superpoint pffnet (128 float32: 512 bytes) pffNet descriptors + SuperPoint keypoints + MAGSAC outlier-filtering N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00011
GeoDesc, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 93.44
Rank: 74/86
0.323
Rank: 68/86
0.759
Rank: 58/86
0.34215
(±0.00054)
Rank: 62/86
120.68
Rank: 83/86
1098.70
Rank: 76/86
4.002
Rank: 67/86
0.552
Rank: 66/86
0.50594
(±0.00073)
Rank: 70/86
0.42405
Rank: 65/86
Challenge organizers (contact) sift-def geodesc (128 float32: 512 bytes) GeoDesc descriptors extracted on SIFT keypoints and DEGENSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-23 is_baseline
Submission ID: 00043
DELF-GLD (128D), PyRANSAC
Size: 512 bytes. Matches: built-in
2036.82 94.70
Rank: 73/86
0.109
Rank: 80/86
0.156
Rank: 82/86
0.07298
(±0.00056)
Rank: 84/86
430.68
Rank: 9/86
2162.72
Rank: 7/86
2.535
Rank: 82/86
0.887
Rank: 83/86
0.13247
(±0.00116)
Rank: 81/86
0.10272
Rank: 82/86
Challenge organizers (contact) delf-gld-2k-128d delf-gld-2k-128d (128 float32: 512 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 128 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with PyRANSAC (DEGENSAC with the degeneracy check turned off). https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-23 is_baseline
Submission ID: ?????
CV-DoG-HardNet8-PTv2
Size: 512 bytes. Matches: built-in
2047.77 153.69
Rank: 31/86
0.339
Rank: 35/86
0.821
Rank: 6/86
0.43869
(±0.00017)
Rank: 23/86
158.63
Rank: 64/86
1221.12
Rank: 63/86
4.350
Rank: 28/86
0.484
Rank: 40/86
0.60206
(±0.00003)
Rank: 44/86
0.52038
Rank: 36/86
Milan Pultar, Dmytro Mishkin, Jiri Matas (contact) sift2k h8e512pt (128 float32: 512 bytes) [sid:00619] HardNet8 with PCA compression, batch sampling from few images N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00616
Sift-Fusion_Avg-NM-Net_End-to-En...
Size: 512 bytes. Matches: custom
1892.70 142.38
Rank: 39/86
0.333
Rank: 57/86
0.801
Rank: 24/86
0.44818
(±0.00000)
Rank: 20/86
146.58
Rank: 74/86
1339.45
Rank: 47/86
4.406
Rank: 26/86
0.456
Rank: 21/86
0.65563
(±0.00209)
Rank: 22/86
0.55190
Rank: 19/86
Chen Zhao (contact) siftdef fusion-avg (128 float32: 512 bytes) SIFT and Fusion_Avg, followed by NM-Net_End-to-End and DEGENSAC. N/A N/A 20-05-31 is_submission, is_challenge_2020
Submission ID: 00704
SOSNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 171.22
Rank: 22/86
0.333
Rank: 42/86
0.804
Rank: 22/86
0.45053
(±0.00053)
Rank: 19/86
194.04
Rank: 44/86
1442.32
Rank: 35/86
4.313
Rank: 38/86
0.467
Rank: 31/86
0.63586
(±0.00246)
Rank: 30/86
0.54320
Rank: 23/86
Challenge organizers (contact) sift-def sosnet-upright (128 float32: 512 bytes) SOSNet descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-06-02 is_baseline
Submission ID: 00507
sift and hardnet512
Size: 2048 bytes. Matches: built-in
2000.00 98.07
Rank: 71/86
0.321
Rank: 76/86
0.778
Rank: 41/86
0.35047
(±0.00046)
Rank: 61/86
153.92
Rank: 68/86
1310.53
Rank: 51/86
4.017
Rank: 65/86
0.515
Rank: 49/86
0.56598
(±0.00185)
Rank: 53/86
0.45822
Rank: 59/86
Ximin Zheng, Sheng He, Hualong Shi (contact) sift8k hardnet512 (512 float32: 2048 bytes) SIFT with 2000 keypoints, hardnet512 with 512 descriptors N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00710
DISK-32D (epi)
Size: 128 bytes. Matches: built-in
2048.00 352.43
Rank: 5/86
0.459
Rank: 3/86
0.826
Rank: 3/86
0.45946
(±0.00023)
Rank: 14/86
484.30
Rank: 2/86
2324.68
Rank: 2/86
5.589
Rank: 1/86
0.427
Rank: 10/86
0.69938
(±0.00192)
Rank: 10/86
0.57942
Rank: 8/86
Under review! NeurIPS anonymous submission, ID 1194 (To be released: 21-09-16) disk-2020-09-15-nms-5-epi-32-save-49-imsize-1024-nump-2048 disk-2020-09-15-nms-5-epi-32-save-49-imsize-1024-nump-2048 (32 float32: 128 bytes) Local feature model learned via policy gradient, using 32D descriptors. Model trained with a cycle-consistency loss and supervised with epipolar constraints. Trained on MegaDepth, removing conflicts with the test data. For inference, images are resized to 1024 pixels on the longest edge, with NMS over a 5x5 window, taking the top 2048 features by score. N/A N/A 20-09-16 is_baseline, is_under_review
Submission ID: 00578
Key.Net-s + DescNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
2033.67 246.62
Rank: 13/86
0.449
Rank: 5/86
0.805
Rank: 21/86
0.45418
(±0.00056)
Rank: 16/86
331.55
Rank: 17/86
1621.68
Rank: 26/86
4.570
Rank: 13/86
0.447
Rank: 16/86
0.67414
(±0.00371)
Rank: 16/86
0.56416
Rank: 15/86
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) keynet-s descnet (128 float32: 512 bytes) N/A N/A 20-05-20 is_submission, is_challenge_2020
Submission ID: 00030
Upright Root-SIFT (OpenCV), DEGE...
Size: 512 bytes. Matches: built-in
1892.72 111.69
Rank: 61/86
0.333
Rank: 38/86
0.781
Rank: 39/86
0.39547
(±0.00027)
Rank: 44/86
201.15
Rank: 41/86
1347.75
Rank: 45/86
4.091
Rank: 56/86
0.522
Rank: 53/86
0.56078
(±0.00264)
Rank: 55/86
0.47812
Rank: 49/86
Challenge organizers (contact) sift-def rootsift-upright (128 float32: 512 bytes) Upright Root-SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00009
GeoDesc (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 116.56
Rank: 55/86
0.333
Rank: 42/86
0.784
Rank: 34/86
0.38660
(±0.00078)
Rank: 48/86
149.66
Rank: 72/86
1237.96
Rank: 60/86
4.219
Rank: 51/86
0.525
Rank: 55/86
0.56687
(±0.00261)
Rank: 51/86
0.47673
Rank: 50/86
Challenge organizers (contact) sift-def geodesc-upright (128 float32: 512 bytes) GeoDesc descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-22 is_baseline
Submission ID: 00701
GeoDesc (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 132.69
Rank: 46/86
0.333
Rank: 42/86
0.798
Rank: 27/86
0.41364
(±0.00033)
Rank: 36/86
161.05
Rank: 63/86
1287.26
Rank: 56/86
4.239
Rank: 46/86
0.502
Rank: 47/86
0.58373
(±0.00402)
Rank: 48/86
0.49868
Rank: 42/86
Challenge organizers (contact) sift-def geodesc-upright (128 float32: 512 bytes) GeoDesc descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-06-01 is_baseline
Submission ID: 00017
ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
2031.83 63.43
Rank: 83/86
0.366
Rank: 22/86
0.523
Rank: 77/86
0.08243
(±0.00058)
Rank: 81/86
120.93
Rank: 82/86
280.38
Rank: 86/86
2.191
Rank: 86/86
0.863
Rank: 79/86
0.02318
(±0.00229)
Rank: 86/86
0.05280
Rank: 86/86
Challenge organizers (contact) orb orb (32 uint8: 32 bytes) ORB with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00028
DELF-GLD (512D), DEGENSAC
Size: 2048 bytes. Matches: built-in
2036.82 98.87
Rank: 70/86
0.109
Rank: 80/86
0.170
Rank: 81/86
0.09156
(±0.00045)
Rank: 79/86
437.64
Rank: 7/86
2296.72
Rank: 3/86
2.567
Rank: 80/86
0.877
Rank: 80/86
0.14991
(±0.00513)
Rank: 79/86
0.12073
Rank: 79/86
Challenge organizers (contact) delf-gld-2k-512d delf-gld-2k-512d (512 float32: 2048 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 512 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with DEGENSAC. https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-23 is_baseline
Submission ID: 00589
SuperPoint-128d + SuperGlue + DE...
Size: 512 bytes. Matches: custom
1973.62 320.46
Rank: 8/86
0.364
Rank: 24/86
0.772
Rank: 49/86
0.55214
(±0.00098)
Rank: 4/86
429.49
Rank: 11/86
2130.73
Rank: 11/86
4.570
Rank: 12/86
0.384
Rank: 4/86
0.75283
(±0.00086)
Rank: 3/86
0.65248
Rank: 3/86
Paul-Edouard Sarlin (contact) superpoint-k2048-nms3-refine2-r1600forcecubic superpoint-down128 (128 float32: 512 bytes) SuperPoint detector (2048 keypoints, NMS with radius 3, refinement, on 1600-pixel images) and descriptor; reduced to 128d with a linear autoencoder. SuperGlue matcher (outdoor model, 150 Sinkhorn iterations). For stereo, DEGENSAC model estimator (1.2 pixels inlier threshold). https://arxiv.org/abs/1911.11763 https://psarlin.com/superglue 20-05-24 is_submission, is_challenge_2020
Submission ID: 00702
L2-Net (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 117.11
Rank: 52/86
0.333
Rank: 42/86
0.808
Rank: 17/86
0.41918
(±0.00059)
Rank: 35/86
179.79
Rank: 50/86
1361.34
Rank: 44/86
4.232
Rank: 47/86
0.481
Rank: 39/86
0.59682
(±0.00079)
Rank: 45/86
0.50800
Rank: 40/86
Challenge organizers (contact) sift-def l2net-upright (128 float32: 512 bytes) L2-Net descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. http://www.nlpr.ia.ac.cn/fanbin/pub/L2-Net_CVPR17.pdf https://github.com/vcg-uvic/image-matching-benchmark-baselines 20-06-01 is_baseline
Submission ID: 00029
SuperPoint (2k features, NMS=4),...
Size: 1024 bytes. Matches: built-in
2048.00 122.69
Rank: 48/86
0.364
Rank: 25/86
0.630
Rank: 71/86
0.28965
(±0.00063)
Rank: 69/86
170.60
Rank: 56/86
1185.38
Rank: 69/86
4.331
Rank: 33/86
0.557
Rank: 67/86
0.54663
(±0.00218)
Rank: 62/86
0.41814
Rank: 67/86
Challenge organizers (contact) superpoint-nms4-r1200 superpoint-nms4-r1200 (256 float32: 1024 bytes) SuperPoint (initial release based on Pytorch). Lowered detection threshold to obtain up to 2k features per image. Using the built-in Non-Maxima Suppression filter at 2 pixels. Images are resized to 1200 pixels on the largest side. https://arxiv.org/abs/1712.07629 https://github.com/MagicLeapResearch/SuperPointPretrainedNetwork 20-04-22 is_baseline
Submission ID: 00603
SuperPoint-128d-masked + SuperGl...
Size: 512 bytes. Matches: custom
2041.09 404.73
Rank: 2/86
0.387
Rank: 16/86
0.774
Rank: 46/86
0.56769
(±0.00034)
Rank: 2/86
415.75
Rank: 13/86
2158.03
Rank: 9/86
4.932
Rank: 5/86
0.357
Rank: 1/86
0.76987
(±0.00122)
Rank: 2/86
0.66878
Rank: 2/86
Paul-Edouard Sarlin (contact) superpoint-k2048-nms3-refine2-r1600forcecubic-masked-d.001 superpoint-down128 (128 float32: 512 bytes) SuperPoint detector (2048 keypoints, NMS with radius 3, confidence threshold 0.001, refinement, on 1600-pixel images). Detections on semantic classes sky and people are removed (segmentation from HFNetV2 trained on MIT ADE20K). SuperPoint descriptor, reduced to 128d with a linear autoencoder. SuperGlue matcher (outdoor model, 150 Sinkhorn iterations). For stereo, DEGENSAC model estimator (1.2 pixel inlier threshold). https://arxiv.org/abs/1911.11763 https://psarlin.com/superglue 20-05-30 is_submission, is_challenge_2020
Submission ID: 00609
Key.Net-s-ref + HyNet w/ DEGENSA...
Size: 512 bytes. Matches: built-in
2032.93 270.04
Rank: 10/86
0.461
Rank: 2/86
0.820
Rank: 8/86
0.46902
(±0.00032)
Rank: 8/86
280.40
Rank: 24/86
1489.62
Rank: 30/86
4.689
Rank: 7/86
0.436
Rank: 13/86
0.68120
(±0.00119)
Rank: 13/86
0.57511
Rank: 12/86
Barroso-Laguna, Axel and Tian, Yurun, Ng, Tony and Mikolajczyk, Krystian (contact) keynet-s-ref hynet (128 float32: 512 bytes) N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00046
D2-Net (single-scale), DEGENSAC
Size: 2048 bytes. Matches: built-in
2038.22 147.53
Rank: 37/86
0.202
Rank: 77/86
0.337
Rank: 78/86
0.16324
(±0.00082)
Rank: 77/86
364.68
Rank: 15/86
1984.98
Rank: 13/86
3.412
Rank: 76/86
0.720
Rank: 76/86
0.36911
(±0.00167)
Rank: 76/86
0.26618
Rank: 76/86
Challenge organizers (contact) d2net-singlescale d2net-singlescale (512 float32: 2048 bytes) D2-Net, single-scale model, up to 2048 features. Trained on the MegaDepth dataset, removing scenes which overlap with the Phototourism test set. Stereo with DEGENSAC and optimal parameters. Parameters tuned separately from the 8k submission. http://openaccess.thecvf.com/content_CVPR_2019/papers/Dusmanu_D2-Net_A_Trainable_CNN_for_Joint_Description_and_Detection_of_CVPR_2019_paper.pdf https://github.com/mihaidusmanu/d2-net 20-04-23 is_baseline
Submission ID: 00036
SOSNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 155.75
Rank: 29/86
0.333
Rank: 42/86
0.794
Rank: 30/86
0.43077
(±0.00017)
Rank: 28/86
183.92
Rank: 47/86
1403.66
Rank: 39/86
4.314
Rank: 36/86
0.472
Rank: 34/86
0.62181
(±0.00161)
Rank: 37/86
0.52629
Rank: 35/86
Challenge organizers (contact) sift-def sosnet-upright (128 float32: 512 bytes) SOSNet descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00615
SIFT-Fusion_Max-NM-Net_End-to-En...
Size: 512 bytes. Matches: custom
1892.70 115.46
Rank: 57/86
0.333
Rank: 57/86
0.783
Rank: 35/86
0.40244
(±0.00000)
Rank: 40/86
119.23
Rank: 85/86
1167.54
Rank: 70/86
4.367
Rank: 27/86
0.489
Rank: 41/86
0.60720
(±0.00114)
Rank: 40/86
0.50482
Rank: 41/86
Chen Zhao (contact) siftdef fusion-max (128 float32: 512 bytes) SIFT and Fusion_Max, followed by NM-Net_End-to-End and DEGENSAC. N/A N/A 20-05-31 is_submission, is_challenge_2020
Submission ID: 00012
GeoDesc (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 144.92
Rank: 38/86
0.333
Rank: 42/86
0.765
Rank: 54/86
0.38124
(±0.00056)
Rank: 54/86
149.66
Rank: 72/86
1237.96
Rank: 60/86
4.219
Rank: 51/86
0.526
Rank: 57/86
0.56687
(±0.00261)
Rank: 51/86
0.47405
Rank: 51/86
Challenge organizers (contact) sift-def geodesc-upright (128 float32: 512 bytes) GeoDesc descriptors extracted on SIFT keypoints with a fixed orientation, and MAGSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-23 is_baseline
Submission ID: 00026
AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
1657.83 69.77
Rank: 82/86
0.355
Rank: 27/86
0.622
Rank: 75/86
0.17632
(±0.00087)
Rank: 76/86
224.40
Rank: 35/86
795.54
Rank: 85/86
3.311
Rank: 77/86
0.819
Rank: 78/86
0.23020
(±0.00466)
Rank: 78/86
0.20326
Rank: 77/86
Challenge organizers (contact) akaze-def akaze (61 uint8: 61 bytes) AKAZE with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00583
R2D2
Size: 512 bytes. Matches: built-in
2048.00 169.38
Rank: 23/86
0.398
Rank: 15/86
0.695
Rank: 67/86
0.33715
(±0.00149)
Rank: 64/86
306.15
Rank: 19/86
1235.83
Rank: 62/86
4.156
Rank: 55/86
0.496
Rank: 43/86
0.60223
(±0.00103)
Rank: 43/86
0.46969
Rank: 56/86
(contact) r2d2-5k-p-aug r2d2-5k-p-aug (128 float32: 512 bytes) N/A N/A 20-05-23 is_submission, is_challenge_2020
Submission ID: 00513
Personal Inplementation SIFT (GP...
Size: 512 bytes. Matches: built-in
1900.60 70.04
Rank: 81/86
0.085
Rank: 86/86
0.002
Rank: 86/86
0.07782
(±0.00022)
Rank: 82/86
173.06
Rank: 52/86
948.78
Rank: 82/86
3.816
Rank: 72/86
0.887
Rank: 84/86
0.09835
(±0.00068)
Rank: 83/86
0.08809
Rank: 83/86
feyman_priv (contact) sift-gpu sift-gpu (128 float32: 512 bytes) SIFT with 2048 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC, and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00013
SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1936.30 76.94
Rank: 79/86
0.323
Rank: 72/86
0.733
Rank: 64/86
0.30766
(±0.00075)
Rank: 68/86
171.92
Rank: 55/86
1059.70
Rank: 80/86
3.687
Rank: 74/86
0.618
Rank: 73/86
0.42476
(±0.00103)
Rank: 73/86
0.36621
Rank: 73/86
Challenge organizers (contact) sift-def sift (128 float32: 512 bytes) SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-23 is_baseline
Submission ID: 00706
LogPolar-Upright w/ DEGENSAC (no...
Size: 512 bytes. Matches: built-in
1892.71 162.20
Rank: 24/86
0.333
Rank: 42/86
0.807
Rank: 19/86
0.45674
(±0.00059)
Rank: 15/86
211.87
Rank: 36/86
1553.36
Rank: 28/86
4.331
Rank: 32/86
0.471
Rank: 33/86
0.63701
(±0.00278)
Rank: 29/86
0.54688
Rank: 21/86
Challenge organizers (contact) sift-def logpolar-upright (128 float32: 512 bytes) Upright LogPolar descriptors on DoG features (OpenCV). Using the built-in matcher: bidirectional filter with the both strategy. DEGENSAC for stereo. FLANN disabled. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-06-02 is_baseline
Submission ID: 00123
KeyNet-HardNet-2k
Size: 512 bytes. Matches: built-in
2048.00 134.39
Rank: 44/86
0.469
Rank: 1/86
0.818
Rank: 9/86
0.32725
(±0.00026)
Rank: 66/86
195.33
Rank: 43/86
1276.30
Rank: 57/86
4.493
Rank: 20/86
0.490
Rank: 42/86
0.61606
(±0.00141)
Rank: 39/86
0.47166
Rank: 53/86
Challenge organizers (contact) keynettuned vlhardnet (128 float32: 512 bytes) KeyNet-HardNet with 2048 features, using the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with RANSAC N/A N/A 20-05-03 is_baseline
Submission ID: 00608
SIFT2k_2048_HardNet64-train-all-...
Size: 512 bytes. Matches: custom
2047.00 245.40
Rank: 14/86
0.346
Rank: 31/86
0.823
Rank: 4/86
0.49219
(±0.00000)
Rank: 6/86
253.37
Rank: 30/86
1984.40
Rank: 14/86
4.607
Rank: 9/86
0.421
Rank: 9/86
0.70020
(±0.00213)
Rank: 9/86
0.59619
Rank: 6/86
Ximin Zheng, Sheng He, Hualong Shi (contact) sift2k hardnet64-train-all-l2-138000-matched (128 float32: 512 bytes) SIFT with 2048 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 138000), FLANN disabled, custom matches N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00044
Upright SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.72 104.22
Rank: 68/86
0.333
Rank: 38/86
0.763
Rank: 56/86
0.37078
(±0.00056)
Rank: 57/86
205.17
Rank: 38/86
1300.49
Rank: 54/86
4.000
Rank: 69/86
0.563
Rank: 68/86
0.52549
(±0.00120)
Rank: 66/86
0.44814
Rank: 62/86
Challenge organizers (contact) sift-def sift-upright (128 float32: 512 bytes) Upright SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC, and setting keypoint orientation to a constant value. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00503
Guided matching hardnet
Size: 512 bytes. Matches: custom
1892.71 149.75
Rank: 35/86
0.333
Rank: 56/86
0.795
Rank: 29/86
0.42159
(±0.00000)
Rank: 34/86
254.13
Rank: 28/86
1754.99
Rank: 21/86
4.283
Rank: 45/86
0.464
Rank: 27/86
0.64144
(±0.00207)
Rank: 28/86
0.53152
Rank: 31/86
(contact) sift upright-hardnet (128 float32: 512 bytes) Sift keypoint ; upright hardnet descriptors ; custom matching: use a deep learning based coarse matcher as a first step. In a second step match the keypoint according to descriptor distances but only for the matches that are close to the coarse match prediction N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00022
L2-Net (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 110.72
Rank: 62/86
0.333
Rank: 42/86
0.801
Rank: 25/86
0.40729
(±0.00097)
Rank: 37/86
171.93
Rank: 53/86
1333.56
Rank: 48/86
4.226
Rank: 49/86
0.497
Rank: 45/86
0.58762
(±0.00228)
Rank: 46/86
0.49746
Rank: 43/86
Challenge organizers (contact) sift-def l2net-upright (128 float32: 512 bytes) L2-Net descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. http://www.nlpr.ia.ac.cn/fanbin/pub/L2-Net_CVPR17.pdf https://github.com/vcg-uvic/image-matching-benchmark-baselines 20-04-23 is_baseline
Submission ID: 00016
LogPolarDesc, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 148.56
Rank: 36/86
0.323
Rank: 62/86
0.752
Rank: 62/86
0.38840
(±0.00064)
Rank: 47/86
162.84
Rank: 61/86
1385.60
Rank: 42/86
4.046
Rank: 59/86
0.516
Rank: 50/86
0.57149
(±0.00344)
Rank: 49/86
0.47994
Rank: 48/86
Challenge organizers (contact) sift-def logpolar (128 float32: 512 bytes) LogPolarDesc descriptors extracted on SIFT keypoints and MAGSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-23 is_baseline
Submission ID: 00622
Sift-HardNet-NM-Net_End-to-End
Size: 512 bytes. Matches: custom
1892.70 152.13
Rank: 33/86
0.333
Rank: 57/86
0.807
Rank: 20/86
0.46620
(±0.00000)
Rank: 10/86
156.54
Rank: 65/86
1403.97
Rank: 38/86
4.425
Rank: 24/86
0.443
Rank: 15/86
0.67160
(±0.00184)
Rank: 17/86
0.56890
Rank: 13/86
Chen Zhao (contact) siftdef hardnet (128 float32: 512 bytes) SIFT and HardNet, followed by NM-Net_End-to-End and DEGENSAC. N/A N/A 20-06-02 is_submission, is_challenge_2020
Submission ID: 00032
Root-SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1936.30 83.46
Rank: 78/86
0.323
Rank: 72/86
0.756
Rank: 61/86
0.33605
(±0.00092)
Rank: 65/86
167.84
Rank: 58/86
1141.88
Rank: 73/86
3.788
Rank: 73/86
0.576
Rank: 71/86
0.46826
(±0.00154)
Rank: 72/86
0.40216
Rank: 70/86
Challenge organizers (contact) sift-def rootsift (128 float32: 512 bytes) Root-SIFT with (up to) 2048 features, using the built-in matcher (bidirectional filter with the both strategy) and DEGENSAC. N/A https://opencv.org 20-04-22 is_baseline
Submission ID: 00040
DELF-GLD (128D), DEGENSAC
Size: 512 bytes. Matches: built-in
2036.82 100.78
Rank: 69/86
0.109
Rank: 80/86
0.151
Rank: 83/86
0.08367
(±0.00047)
Rank: 80/86
430.68
Rank: 9/86
2162.72
Rank: 7/86
2.535
Rank: 82/86
0.880
Rank: 82/86
0.13247
(±0.00116)
Rank: 81/86
0.10807
Rank: 81/86
Challenge organizers (contact) delf-gld-2k-128d delf-gld-2k-128d (128 float32: 512 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 128 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with DEGENSAC. https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-22 is_baseline
Submission ID: 00515
SuperPoint + SuperGlue + DEGENSA...
Size: 1024 bytes. Matches: custom
1973.62 320.50
Rank: 7/86
0.364
Rank: 23/86
0.772
Rank: 48/86
0.55234
(±0.00024)
Rank: 3/86
429.48
Rank: 12/86
2131.07
Rank: 10/86
4.572
Rank: 11/86
0.373
Rank: 3/86
0.75213
(±0.00345)
Rank: 4/86
0.65224
Rank: 4/86
Paul-Edouard Sarlin (contact) superpoint-nms3-refine2-r1600forcecubic superpoint (256 float32: 1024 bytes) SuperPoint detector and descriptor (2048 keypoints, NMS with radius 3, refinement, on 1600-pixel images). SuperGlue matcher (outdoor model, 150 Sinkhorn iterations). For stereo, DEGENSAC model estimator (1.2 pixels inlier threshold). https://arxiv.org/abs/1911.11763 https://github.com/magicleap/SuperGluePretrainedNetwork 20-04-24 is_submission, is_challenge_2020
Submission ID: 00605
guided-hardnet-epoch2-v2
Size: 512 bytes. Matches: custom
2047.76 220.88
Rank: 16/86
0.347
Rank: 28/86
0.778
Rank: 42/86
0.46469
(±0.00000)
Rank: 11/86
226.84
Rank: 34/86
1827.60
Rank: 18/86
4.512
Rank: 17/86
0.427
Rank: 11/86
0.69306
(±0.00160)
Rank: 12/86
0.57888
Rank: 11/86
Chen Shen, Zhipeng Wang, Jun Zhang, Zhongkun Chen, Zhiwei Ruan, Jingchao Zhou, Pengfei Xu (contact) sift2k hardnet-epoch2 (128 float32: 512 bytes) sift and hardnet with 2k features, using the oanet and guided matching and setting keypoint orientation to a constant value to increase performance. N/A N/A 20-05-30 is_submission, is_challenge_2020
Submission ID: 00027
DELF-GLD (32D), PyRANSAC
Size: 128 bytes. Matches: built-in
2036.82 91.18
Rank: 76/86
0.109
Rank: 80/86
0.122
Rank: 84/86
0.06197
(±0.00032)
Rank: 86/86
453.19
Rank: 3/86
1805.48
Rank: 19/86
2.425
Rank: 84/86
0.916
Rank: 85/86
0.09163
(±0.00140)
Rank: 84/86
0.07680
Rank: 85/86
Challenge organizers (contact) delf-gld-2k-32d delf-gld-2k-32d (32 float32: 128 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 32 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with PyRANSAC (DEGENSAC with the degeneracy check turned off). https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-22 is_baseline
Submission ID: 00037
GeoDesc, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 115.17
Rank: 58/86
0.323
Rank: 68/86
0.742
Rank: 63/86
0.34024
(±0.00050)
Rank: 63/86
120.68
Rank: 83/86
1098.70
Rank: 76/86
4.002
Rank: 67/86
0.551
Rank: 65/86
0.50594
(±0.00073)
Rank: 70/86
0.42309
Rank: 66/86
Challenge organizers (contact) sift-def geodesc (128 float32: 512 bytes) GeoDesc descriptors extracted on SIFT keypoints and MAGSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1807.06294 https://github.com/lzx551402/geodesc 20-04-23 is_baseline
Submission ID: 00023
LogPolarDesc, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 116.83
Rank: 53/86
0.323
Rank: 62/86
0.770
Rank: 50/86
0.39335
(±0.00060)
Rank: 45/86
162.84
Rank: 61/86
1385.60
Rank: 42/86
4.046
Rank: 59/86
0.519
Rank: 52/86
0.57149
(±0.00344)
Rank: 49/86
0.48242
Rank: 45/86
Challenge organizers (contact) sift-def logpolar (128 float32: 512 bytes) LogPolarDesc descriptors extracted on SIFT keypoints and DEGENSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1908.05547 https://github.com/cvlab-epfl/log-polar-descriptors 20-04-22 is_baseline
Submission ID: 00564
SIFT2k_2000_HardNet64_train_all_...
Size: 512 bytes. Matches: built-in
1999.83 156.93
Rank: 28/86
0.336
Rank: 36/86
0.814
Rank: 11/86
0.42959
(±0.00060)
Rank: 29/86
244.13
Rank: 33/86
1646.76
Rank: 25/86
4.334
Rank: 31/86
0.459
Rank: 24/86
0.64407
(±0.00339)
Rank: 25/86
0.53683
Rank: 28/86
Ximin Zheng, Sheng He, Hualong Shi (contact) sift2k sift2k-2000-hardnet64-train-all-l2-138000 (128 float32: 512 bytes) SIFT with 2000 keypoints(scale 12), hardnet64 with 128 descriptors(trained with l2 loss and step 138000), FLANN disabled N/A N/A 20-05-13 is_submission, is_challenge_2020
Submission ID: 00025
DELF-GLD (32D), DEGENSAC
Size: 128 bytes. Matches: built-in
2036.82 96.56
Rank: 72/86
0.109
Rank: 80/86
0.119
Rank: 85/86
0.06573
(±0.00030)
Rank: 85/86
453.19
Rank: 3/86
1805.48
Rank: 19/86
2.425
Rank: 84/86
0.920
Rank: 86/86
0.09163
(±0.00140)
Rank: 84/86
0.07868
Rank: 84/86
Challenge organizers (contact) delf-gld-2k-32d delf-gld-2k-32d (32 float32: 128 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 32 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with DEGENSAC. https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-22 is_baseline
Submission ID: 00512
Hardnet with DEGENSAC for stereo
Size: 512 bytes. Matches: built-in
1999.50 51.91
Rank: 86/86
0.172
Rank: 78/86
0.558
Rank: 76/86
0.20771
(±0.00049)
Rank: 74/86
181.48
Rank: 49/86
1692.63
Rank: 23/86
3.050
Rank: 78/86
0.695
Rank: 74/86
0.38463
(±0.00275)
Rank: 74/86
0.29617
Rank: 74/86
Vu Trung Nghia & Nguyen Trung Hieu (contact) siftdef hardnet-64 (128 float32: 512 bytes) Using hardnet network to embed a patch (32 x 32) to a 128 (float32) dimensions vector, For stereo, we use DEGENSAC (Chum et al, CVPR'05) with optimal settings N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: 00527
SEKD
Size: 512 bytes. Matches: built-in
1786.69 106.34
Rank: 65/86
0.383
Rank: 19/86
0.759
Rank: 59/86
0.38886
(±0.00013)
Rank: 46/86
150.11
Rank: 71/86
1063.08
Rank: 79/86
4.575
Rank: 10/86
0.465
Rank: 29/86
0.63303
(±0.00244)
Rank: 32/86
0.51095
Rank: 39/86
Yafei Song, Ling Cai, Mingyang Li (contact) sekd sekd (128 float32: 512 bytes) The name of our method is Self-Evolving Keypoint Detection and Description (SEKD). Now, the SEKD model is trained only using COCO test images. In this submission each image has up to 2048 SEKD keypoints, and 128-dim float descriptor. We use the built-in matcher (bidirectional filter with the both strategy, optimal inlier and ratio test thresholds) with DEGENSAC. N/A N/A 20-04-27 is_submission, is_challenge_2020
Submission ID: 00031
SuperPoint (2k features, NMS=2),...
Size: 1024 bytes. Matches: built-in
2048.00 105.52
Rank: 66/86
0.376
Rank: 20/86
0.624
Rank: 74/86
0.27954
(±0.00073)
Rank: 71/86
130.52
Rank: 81/86
1002.03
Rank: 81/86
4.211
Rank: 53/86
0.593
Rank: 72/86
0.51061
(±0.00378)
Rank: 69/86
0.39508
Rank: 71/86
Challenge organizers (contact) superpoint-nms2-r1200 superpoint-nms2-r1200 (256 float32: 1024 bytes) SuperPoint (initial release based on Pytorch). Lowered detection threshold to obtain up to 2k features per image. Using the built-in Non-Maxima Suppression filter at 2 pixels. Images are resized to 1200 pixels on the largest side. https://arxiv.org/abs/1712.07629 https://github.com/MagicLeapResearch/SuperPointPretrainedNetwork 20-04-22 is_baseline
Submission ID: 00711
DISK-32D (depth)
Size: 128 bytes. Matches: built-in
2048.00 341.83
Rank: 6/86
0.448
Rank: 6/86
0.842
Rank: 2/86
0.47005
(±0.00037)
Rank: 7/86
441.16
Rank: 6/86
2231.55
Rank: 6/86
5.524
Rank: 3/86
0.415
Rank: 6/86
0.70220
(±0.00208)
Rank: 8/86
0.58613
Rank: 7/86
Under review! NeurIPS anonymous submission, ID 1194 (To be released: 21-09-16) disk-2020-09-15-nms-7-depth-32-save-46-imsize-1024-nump-2048 disk-2020-09-15-nms-7-depth-32-save-46-imsize-1024-nump-2048 (32 float32: 128 bytes) Local feature model learned via policy gradient, using 32D descriptors. Model trained with a cycle-consistency loss and supervised with depth-based constraints. Trained on MegaDepth, removing conflicts with the test data. For inference, images are resized to 1024 pixels on the longest edge, with NMS over a 7x7 window, taking the top 2048 features by score. N/A N/A 20-09-16 is_baseline, is_under_review
Submission ID: 00612
SuperPoint-128d-adapt + SuperGlu...
Size: 512 bytes. Matches: custom
2048.00 441.49
Rank: 1/86
0.407
Rank: 13/86
0.789
Rank: 32/86
0.59034
(±0.00050)
Rank: 1/86
452.99
Rank: 5/86
2245.43
Rank: 5/86
5.092
Rank: 4/86
0.358
Rank: 2/86
0.77337
(±0.00167)
Rank: 1/86
0.68186
Rank: 1/86
Paul-Edouard Sarlin (contact) superpoint-k2048-nms4-refine2-r1600forcecubic-masked-d.001-adapt50 superpoint-down128 (128 float32: 512 bytes) SuperPoint detector (2048 keypoints, NMS with radius 4, confidence threshold 0.001, refinement, on 1600-pixel images). The detection heatmap is improved with test-time homographic adaptation (50 iterations), and detections on semantic classes sky and people are removed (segmentation from HFNetV2 trained on MIT ADE20K). SuperPoint descriptor, reduced to 128d with a linear autoencoder. SuperGlue matcher (outdoor model, 150 Sinkhorn iterations). For stereo, DEGENSAC model estimator (1.1 pixel inlier threshold). https://arxiv.org/abs/1911.11763 https://psarlin.com/superglue 20-05-31 is_submission, is_challenge_2020
Submission ID: 00021
DELF-GLD (512D), PyRANSAC
Size: 2048 bytes. Matches: built-in
2036.82 92.96
Rank: 75/86
0.109
Rank: 80/86
0.175
Rank: 80/86
0.07687
(±0.00010)
Rank: 83/86
437.64
Rank: 7/86
2296.72
Rank: 3/86
2.567
Rank: 80/86
0.878
Rank: 81/86
0.14991
(±0.00513)
Rank: 79/86
0.11339
Rank: 80/86
Challenge organizers (contact) delf-gld-2k-512d delf-gld-2k-512d (512 float32: 2048 bytes) DELF-GLD, with up to 2k features. Descriptors are cropped to 512 dimensions with PCA. Re-using optimal parameters for the (default) 40D models. Stereo with PyRANSAC (DEGENSAC with the degeneracy check turned off). https://arxiv.org/abs/1812.01584 https://github.com/tensorflow/models/tree/master/research/delf 20-04-22 is_baseline
Submission ID: 00586
KeyNet+HardNet+NM-Net_v2
Size: 512 bytes. Matches: custom
2042.89 160.25
Rank: 26/86
0.435
Rank: 11/86
0.821
Rank: 5/86
0.40264
(±0.00000)
Rank: 39/86
165.01
Rank: 60/86
1134.69
Rank: 74/86
4.720
Rank: 6/86
0.448
Rank: 18/86
0.67488
(±0.00197)
Rank: 15/86
0.53876
Rank: 25/86
Chen Zhao (contact) keynet hardnet (128 float32: 512 bytes) KeyNet and HardNet, followed by NM-Net_v2 and DEGENSAC. N/A N/A 20-05-23 is_submission, is_challenge_2020
Submission ID: 00020
D2-Net (multi-scale), DEGENSAC
Size: 2048 bytes. Matches: built-in
2045.55 118.41
Rank: 50/86
0.168
Rank: 79/86
0.293
Rank: 79/86
0.12270
(±0.00048)
Rank: 78/86
286.10
Rank: 22/86
1999.37
Rank: 12/86
3.013
Rank: 79/86
0.774
Rank: 77/86
0.28297
(±0.00254)
Rank: 77/86
0.20283
Rank: 78/86
Challenge organizers (contact) d2net-multiscale d2net-multiscale (512 float32: 2048 bytes) D2-Net, multi-scale model, up to 2048 features. Trained on the MegaDepth dataset, removing scenes which overlap with the Phototourism test set. Stereo with DEGENSAC and optimal parameters. Parameters tuned separately from the 8k submission. http://openaccess.thecvf.com/content_CVPR_2019/papers/Dusmanu_D2-Net_A_Trainable_CNN_for_Joint_Description_and_Detection_of_CVPR_2019_paper.pdf https://github.com/mihaidusmanu/d2-net 20-04-23 is_baseline
Submission ID: 00585
NM-Net_v2
Size: 512 bytes. Matches: custom
1892.70 151.46
Rank: 34/86
0.333
Rank: 57/86
0.809
Rank: 15/86
0.46620
(±0.00000)
Rank: 9/86
155.91
Rank: 67/86
1392.47
Rank: 41/86
4.424
Rank: 25/86
0.459
Rank: 23/86
0.66837
(±0.00125)
Rank: 18/86
0.56729
Rank: 14/86
Chen Zhao (contact) siftdef hardnet (128 float32: 512 bytes) SIFT and HardNet, followed by NM-Net_v2 and DEGENSAC. N/A N/A 20-05-23 is_submission, is_challenge_2020
Submission ID: 00517
retrained Hardnet with MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 133.39
Rank: 45/86
0.333
Rank: 42/86
0.808
Rank: 18/86
0.40042
(±0.00077)
Rank: 41/86
210.10
Rank: 37/86
1480.67
Rank: 31/86
4.226
Rank: 48/86
0.479
Rank: 38/86
0.63030
(±0.00242)
Rank: 33/86
0.51536
Rank: 37/86
Vu Trung Nghia and Nguyen Trung Hieu (contact) sift-upright hardnet (128 float32: 512 bytes) Using SIFT upright and retrained Hardnet for keypoints and descriptors. For stereo, we use DEGENSAC Chum et al, CVPR05 with optimal settings N/A N/A 20-04-24 is_submission, is_challenge_2020
Submission ID: 00509
Upright-Sift + X-Net-lib w/ DEGE...
Size: 512 bytes. Matches: built-in
1892.70 116.55
Rank: 56/86
0.333
Rank: 61/86
0.821
Rank: 7/86
0.42382
(±0.00089)
Rank: 31/86
168.67
Rank: 57/86
1319.63
Rank: 50/86
4.303
Rank: 41/86
0.468
Rank: 32/86
0.62953
(±0.00216)
Rank: 34/86
0.52668
Rank: 34/86
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) sift-def x-net-lib-upright-no-dups (128 float32: 512 bytes) N/A N/A 20-04-25 is_submission, is_challenge_2020
Submission ID: ?????
CV-DoG-HardNet8-PT
Size: 512 bytes. Matches: built-in
2048.00 114.84
Rank: 59/86
0.321
Rank: 75/86
0.790
Rank: 31/86
0.38200
(±0.00081)
Rank: 53/86
117.84
Rank: 86/86
1067.13
Rank: 78/86
4.051
Rank: 58/86
0.531
Rank: 58/86
0.53035
(±0.00531)
Rank: 64/86
0.45617
Rank: 60/86
Milan Pultar, Dmytro Mishkin, Jiri Matas (contact) sift2k h8e512pt (128 float32: 512 bytes) [sid:00593] HardNet8 with PCA compression N/A N/A 20-05-27 is_submission, is_challenge_2020
Submission ID: 00003
HardNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 141.66
Rank: 40/86
0.333
Rank: 42/86
0.803
Rank: 23/86
0.44235
(±0.00127)
Rank: 22/86
193.09
Rank: 45/86
1435.93
Rank: 36/86
4.302
Rank: 42/86
0.474
Rank: 36/86
0.62275
(±0.00423)
Rank: 35/86
0.53255
Rank: 30/86
Challenge organizers (contact) sift-def hardnet-upright (128 float32: 512 bytes) HardNet descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00035
HardNet (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 181.81
Rank: 18/86
0.333
Rank: 42/86
0.781
Rank: 38/86
0.43816
(±0.00058)
Rank: 24/86
193.09
Rank: 45/86
1435.93
Rank: 36/86
4.302
Rank: 42/86
0.473
Rank: 35/86
0.62275
(±0.00423)
Rank: 35/86
0.53045
Rank: 32/86
Challenge organizers (contact) sift-def hardnet-upright (128 float32: 512 bytes) HardNet descriptors extracted on SIFT keypoints with a fixed orientation, and MAGSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-23 is_baseline
Submission ID: 00708
DISK (LCC/depth)
Size: 512 bytes. Matches: built-in
2048.00 404.19
Rank: 3/86
0.448
Rank: 7/86
0.852
Rank: 1/86
0.51315
(±0.00028)
Rank: 5/86
527.48
Rank: 1/86
2428.04
Rank: 1/86
5.545
Rank: 2/86
0.410
Rank: 5/86
0.72705
(±0.00094)
Rank: 5/86
0.62010
Rank: 5/86
Under review! NeurIPS anonymous submission, ID 1194 (To be released: 21-06-03) disk-cc-continued-20-imsize-1024-nms-3-nump-2048 disk-cc-continued-20-imsize-1024-nms-3-nump-2048 (128 float32: 512 bytes) Local feature model learned via policy gradient. Model trained with a cycle-consistency loss and supervised with depth. Trained on MegaDepth, removing conflicts with the test data. For inference, images are resized to 1024 pixels on the longest edge, with NMS over a 3x3 window. We take the top 2048 features by score. N/A N/A 20-06-03 is_submission, is_challenge_2020, is_under_review
Submission ID: 00004
SOSNet (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 176.54
Rank: 19/86
0.333
Rank: 42/86
0.782
Rank: 36/86
0.43803
(±0.00092)
Rank: 25/86
183.92
Rank: 47/86
1403.66
Rank: 39/86
4.314
Rank: 36/86
0.474
Rank: 37/86
0.62181
(±0.00161)
Rank: 37/86
0.52992
Rank: 33/86
Challenge organizers (contact) sift-def sosnet-upright (128 float32: 512 bytes) SOSNet descriptors extracted on SIFT keypoints with a fixed orientation, and MAGSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00018
R2D2 (r2d2-wasf-n16), DEGENSAC
Size: 512 bytes. Matches: built-in
2048.00 176.49
Rank: 20/86
0.422
Rank: 12/86
0.725
Rank: 65/86
0.35783
(±0.00061)
Rank: 59/86
262.02
Rank: 26/86
1188.11
Rank: 68/86
4.299
Rank: 44/86
0.499
Rank: 46/86
0.60639
(±0.00124)
Rank: 41/86
0.48211
Rank: 46/86
Challenge organizers (contact) r2d2-wasf-n16 r2d2-wasf-n16 (128 float32: 512 bytes) R2D2 (model r2d2-wasf-n16) with 2k features, using the built-in matcher (bidirectional filter with the both strategy) and no ratio test, and DEGENSAC with optimal parameters for stereo. N/A https://opencv.org 20-04-23 is_baseline
Submission ID: 00014
L2-Net, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 107.38
Rank: 64/86
0.323
Rank: 68/86
0.766
Rank: 52/86
0.35649
(±0.00110)
Rank: 60/86
138.97
Rank: 77/86
1211.49
Rank: 64/86
3.948
Rank: 70/86
0.538
Rank: 63/86
0.52019
(±0.00120)
Rank: 67/86
0.43834
Rank: 64/86
Challenge organizers (contact) sift-def l2net (128 float32: 512 bytes) L2-Net descriptors extracted on SIFT keypoints and MAGSAC for stereo. Please refer to the baselines repository (linked) for details. http://www.nlpr.ia.ac.cn/fanbin/pub/L2-Net_CVPR17.pdf https://github.com/vcg-uvic/image-matching-benchmark-baselines 20-04-23 is_baseline
Submission ID: 00700
HardNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 152.69
Rank: 32/86
0.333
Rank: 42/86
0.812
Rank: 14/86
0.46093
(±0.00080)
Rank: 13/86
201.26
Rank: 40/86
1467.86
Rank: 34/86
4.311
Rank: 39/86
0.466
Rank: 30/86
0.63544
(±0.00324)
Rank: 31/86
0.54819
Rank: 20/86
Challenge organizers (contact) sift-def hardnet-upright (128 float32: 512 bytes) HardNet descriptors extracted on SIFT keypoints with a fixed orientation, and DEGENSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-06-01 is_baseline
Submission ID: 00033
SOSNet, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 119.91
Rank: 49/86
0.323
Rank: 62/86
0.769
Rank: 51/86
0.37632
(±0.00056)
Rank: 55/86
145.47
Rank: 75/86
1271.41
Rank: 58/86
4.045
Rank: 61/86
0.524
Rank: 54/86
0.56072
(±0.00077)
Rank: 56/86
0.46852
Rank: 57/86
Challenge organizers (contact) sift-def sosnet (128 float32: 512 bytes) SOSNet descriptors extracted on SIFT keypoints and DEGENSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1904.05019 https://github.com/yuruntian/SOSNet 20-04-22 is_baseline
Submission ID: 00039
HardNet, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 138.43
Rank: 42/86
0.323
Rank: 62/86
0.757
Rank: 60/86
0.38490
(±0.00120)
Rank: 52/86
153.29
Rank: 69/86
1306.64
Rank: 52/86
4.026
Rank: 63/86
0.537
Rank: 62/86
0.55573
(±0.00175)
Rank: 60/86
0.47032
Rank: 55/86
Challenge organizers (contact) sift-def hardnet (128 float32: 512 bytes) HardNet descriptors extracted on SIFT keypoints and MAGSAC for stereo. Please refer to the baselines repository (linked) for details. https://arxiv.org/abs/1705.10872 https://github.com/DagnyT/hardnet 20-04-22 is_baseline
Submission ID: 00582
R2D2
Size: 512 bytes. Matches: built-in
2048.00 155.21
Rank: 30/86
0.403
Rank: 14/86
0.709
Rank: 66/86
0.32501
(±0.00085)
Rank: 67/86
285.11
Rank: 23/86
1162.94
Rank: 72/86
4.161
Rank: 54/86
0.508
Rank: 48/86
0.60276
(±0.00062)
Rank: 42/86
0.46389
Rank: 58/86
(contact) r2d2-5k-p r2d2-5k-p (128 float32: 512 bytes) N/A N/A 20-05-23 is_submission, is_challenge_2020
Submission ID: 00617
pffNet + SuperPoint + DEGENSAC
Size: 512 bytes. Matches: built-in
1267.22 55.58
Rank: 85/86
0.341
Rank: 32/86
0.629
Rank: 72/86
0.25514
(±0.00059)
Rank: 72/86
131.91
Rank: 79/86
893.88
Rank: 83/86
4.339
Rank: 29/86
0.534
Rank: 60/86
0.55935
(±0.00143)
Rank: 58/86
0.40724
Rank: 69/86
Jongmin Lee, Seungwook Kim, Yoonwoo Jeong (contact) superpoint pffnet (128 float32: 512 bytes) pffNet descriptors+ SuperPoint keypoints + DEGENSAC outlier-filtering N/A N/A 20-06-01 is_submission, is_challenge_2020
Submission ID: 00008
L2-Net (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 138.44
Rank: 41/86
0.333
Rank: 42/86
0.786
Rank: 33/86
0.39980
(±0.00020)
Rank: 42/86
171.93
Rank: 53/86
1333.56
Rank: 48/86
4.226
Rank: 49/86
0.496
Rank: 44/86
0.58762
(±0.00228)
Rank: 46/86
0.49371
Rank: 44/86
Challenge organizers (contact) sift-def l2net-upright (128 float32: 512 bytes) L2-Net descriptors extracted on SIFT keypoints with a fixed orientation, and MAGSAC for stereo. Re-using parameters found for the orientation-sensitive variant. Please refer to the baselines repository (linked) for details. http://www.nlpr.ia.ac.cn/fanbin/pub/L2-Net_CVPR17.pdf https://github.com/vcg-uvic/image-matching-benchmark-baselines 20-04-22 is_baseline
Submission ID: 00531
Key.Net + X-Net w/ DEGENSAC
Size: 512 bytes. Matches: built-in
2040.08 173.65
Rank: 21/86
0.447
Rank: 8/86
0.797
Rank: 28/86
0.38608
(±0.00068)
Rank: 50/86
276.91
Rank: 25/86
1468.50
Rank: 33/86
4.447
Rank: 21/86
0.463
Rank: 26/86
0.64369
(±0.00081)
Rank: 27/86
0.51489
Rank: 38/86
Barroso-Laguna, Axel and Tian, Yurun and Ng, Tony (contact) keynet x-net (128 float32: 512 bytes) N/A N/A 20-04-24 is_submission, is_challenge_2020