IMW 2020: Leaderboard

Current version: 20210086 (2020-05-28, 03:35 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
[sid:00078] SOSNet-Upright w/ MAGSAC (no...
Size: 512 bytes. Matches: built-in
7829.63 693.71
Rank: 14/113
0.486
Rank: 36/113
0.855
Rank: 25/113
0.56626
(±0.00020)
Rank: 14/113
600.23
Rank: 23/113
4765.43
Rank: 16/113
4.584
Rank: 18/113
0.403
Rank: 17/113
0.73027
(±0.00284)
Rank: 15/113
0.64826
Rank: 16/113
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
[sid:00597] SIFT8k-HardNet64
Size: 512 bytes. Matches: built-in
6589.88 390.09
Rank: 70/113
0.467
Rank: 97/113
0.852
Rank: 30/113
0.52511
(±0.00056)
Rank: 42/113
420.42
Rank: 63/113
3595.93
Rank: 69/113
4.496
Rank: 37/113
0.417
Rank: 32/113
0.70738
(±0.00218)
Rank: 38/113
0.61625
Rank: 35/113
Anonymous (released after deadline) sift8k hardnet (128 float32: 512 bytes) Anonymous (released after deadline) https://arxiv.org/abs/1705.10872 Anonymous (released after deadline) 20-05-27 is_baseline
[sid:00042] ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
7150.21 165.24
Rank: 110/113
0.514
Rank: 15/113
0.653
Rank: 101/113
0.16159
(±0.00090)
Rank: 104/113
910.31
Rank: 11/113
1423.38
Rank: 112/113
2.722
Rank: 112/113
0.897
Rank: 113/113
0.08054
(±0.00242)
Rank: 113/113
0.12106
Rank: 113/113
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
[sid:00112] VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 209.62
Rank: 106/113
0.547
Rank: 11/113
0.823
Rank: 76/113
0.36954
(±0.00081)
Rank: 96/113
347.39
Rank: 95/113
3209.10
Rank: 89/113
4.126
Rank: 91/113
0.517
Rank: 100/113
0.58657
(±0.00120)
Rank: 99/113
0.47806
Rank: 98/113
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
[sid:00508] modified-r2d2
Size: 512 bytes. Matches: built-in
7791.82 163.86
Rank: 111/113
0.517
Rank: 14/113
0.785
Rank: 91/113
0.32815
(±0.00037)
Rank: 97/113
328.39
Rank: 101/113
2811.27
Rank: 100/113
4.250
Rank: 78/113
0.488
Rank: 92/113
0.62899
(±0.00193)
Rank: 88/113
0.47857
Rank: 97/113
Anonymous (released after deadline) modified-r2d2 modified-r2d2 (128 float32: 512 bytes) Anonymous (released after deadline) https://europe.naverlabs.com/wp-content/uploads/2019/09/R2D2-Repeatable-and-Reliable-Detector-and-Descriptor-2.pdf Anonymous (released after deadline) 20-04-25 is_submission
[sid:00536] HarrisZ + DeepOrientation + ...
Size: 512 bytes. Matches: custom
2410.18 1181.11
Rank: 2/113
0.368
Rank: 104/113
0.341
Rank: 111/113
0.00607
(±0.00000)
Rank: 112/113
1191.31
Rank: 3/113
2215.65
Rank: 104/113
4.014
Rank: 102/113
0.524
Rank: 102/113
0.58741
(±0.00250)
Rank: 98/113
0.29674
Rank: 106/113
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
[sid:00019] Upright Root-SIFT (OpenCV), ...
Size: 512 bytes. Matches: built-in
7829.24 363.31
Rank: 80/113
0.487
Rank: 21/113
0.844
Rank: 46/113
0.51009
(±0.00033)
Rank: 58/113
548.18
Rank: 28/113
4406.41
Rank: 34/113
4.385
Rank: 47/113
0.442
Rank: 58/113
0.68859
(±0.00254)
Rank: 59/113
0.59934
Rank: 57/113
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
[sid:00085] LogPolar-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 623.53
Rank: 17/113
0.486
Rank: 36/113
0.830
Rank: 65/113
0.51287
(±0.00038)
Rank: 55/113
483.93
Rank: 48/113
4405.57
Rank: 35/113
4.542
Rank: 29/113
0.423
Rank: 42/113
0.70680
(±0.00247)
Rank: 39/113
0.60983
Rank: 49/113
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
[sid:00571] ContextDesc Upright + OANetV...
Size: 512 bytes. Matches: custom
7830.09 843.99
Rank: 8/113
0.487
Rank: 23/113
0.775
Rank: 92/113
0.53810
(±0.00000)
Rank: 33/113
855.38
Rank: 12/113
6561.52
Rank: 2/113
4.522
Rank: 31/113
0.370
Rank: 4/113
0.76702
(±0.00081)
Rank: 2/113
0.65256
Rank: 13/113
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_baseline
[sid:00573] Guided-HardNet-qht
Size: 512 bytes. Matches: custom
7829.63 902.48
Rank: 6/113
0.486
Rank: 36/113
0.774
Rank: 93/113
0.55224
(±0.00000)
Rank: 21/113
538.28
Rank: 30/113
4693.85
Rank: 19/113
4.650
Rank: 5/113
0.381
Rank: 7/113
0.75930
(±0.00177)
Rank: 4/113
0.65577
Rank: 11/113
Anonymous (released after deadline) sift8k hardnet-qht (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-23 is_baseline
[sid:00010] AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
7857.11 251.77
Rank: 101/113
0.553
Rank: 8/113
0.735
Rank: 99/113
0.30717
(±0.00122)
Rank: 100/113
479.55
Rank: 52/113
2778.68
Rank: 101/113
3.393
Rank: 110/113
0.737
Rank: 112/113
0.36048
(±0.00382)
Rank: 111/113
0.33383
Rank: 102/113
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
[sid:00068] GeoDesc-Upright w/ DEGENSAC ...
Size: 512 bytes. Matches: built-in
7829.63 418.12
Rank: 63/113
0.486
Rank: 36/113
0.869
Rank: 12/113
0.52674
(±0.00049)
Rank: 40/113
458.60
Rank: 53/113
4146.83
Rank: 46/113
4.412
Rank: 43/113
0.431
Rank: 50/113
0.70442
(±0.00189)
Rank: 42/113
0.61558
Rank: 36/113
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
[sid:00532] affnet-hardnet
Size: 512 bytes. Matches: built-in
7925.53 231.83
Rank: 103/113
0.452
Rank: 100/113
0.649
Rank: 103/113
0.31958
(±0.00055)
Rank: 98/113
243.28
Rank: 110/113
3155.72
Rank: 92/113
3.595
Rank: 107/113
0.554
Rank: 103/113
0.52578
(±0.00343)
Rank: 103/113
0.42268
Rank: 100/113
Anonymous (released after deadline) affnet hardnet (128 float32: 512 bytes) Anonymous (released after deadline) https://arxiv.org/abs/1711.06704 Anonymous (released after deadline) 20-04-24 is_submission
[sid:00071] L2-Net w/ DEGENSAC (no FLANN...
Size: 512 bytes. Matches: built-in
7861.11 374.25
Rank: 75/113
0.472
Rank: 77/113
0.852
Rank: 33/113
0.52953
(±0.00043)
Rank: 39/113
339.00
Rank: 99/113
3424.94
Rank: 78/113
4.206
Rank: 83/113
0.452
Rank: 72/113
0.66445
(±0.00213)
Rank: 73/113
0.59699
Rank: 60/113
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
[sid:00024] SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
7860.73 244.12
Rank: 102/113
0.472
Rank: 72/113
0.824
Rank: 74/113
0.45426
(±0.00097)
Rank: 87/113
418.86
Rank: 64/113
3515.63
Rank: 75/113
4.001
Rank: 103/113
0.502
Rank: 95/113
0.60193
(±0.00185)
Rank: 96/113
0.52810
Rank: 88/113
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
[sid:00084] LogPolar-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 468.77
Rank: 49/113
0.486
Rank: 36/113
0.852
Rank: 31/113
0.52316
(±0.00034)
Rank: 46/113
483.93
Rank: 48/113
4405.57
Rank: 35/113
4.542
Rank: 29/113
0.422
Rank: 39/113
0.70680
(±0.00247)
Rank: 39/113
0.61498
Rank: 37/113
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
[sid:00506] sift and hardnet64
Size: 512 bytes. Matches: built-in
7861.62 366.53
Rank: 79/113
0.472
Rank: 75/113
0.849
Rank: 36/113
0.52085
(±0.00009)
Rank: 47/113
585.33
Rank: 25/113
4957.86
Rank: 15/113
4.294
Rank: 70/113
0.419
Rank: 36/113
0.70472
(±0.00257)
Rank: 41/113
0.61279
Rank: 42/113
Anonymous (released after deadline) sift8k hardnet64 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00100] KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 456.82
Rank: 52/113
0.582
Rank: 1/113
0.838
Rank: 52/113
0.39971
(±0.00147)
Rank: 92/113
356.21
Rank: 86/113
3366.01
Rank: 81/113
4.319
Rank: 64/113
0.462
Rank: 80/113
0.64829
(±0.00174)
Rank: 81/113
0.52400
Rank: 89/113
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
[sid:00569] Upright-SIFT-HardNet
Size: 512 bytes. Matches: built-in
6667.86 384.82
Rank: 71/113
0.491
Rank: 17/113
0.844
Rank: 44/113
0.51560
(±0.00043)
Rank: 52/113
397.09
Rank: 73/113
3417.56
Rank: 80/113
4.479
Rank: 39/113
0.448
Rank: 71/113
0.68803
(±0.00532)
Rank: 60/113
0.60181
Rank: 55/113
Anonymous (released after deadline) sift8k hardnet (128 float32: 512 bytes) Anonymous (released after deadline) https://arxiv.org/abs/1705.10872 Anonymous (released after deadline) 20-05-20 is_baseline
[sid:00066] SOSNet-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 549.86
Rank: 32/113
0.486
Rank: 36/113
0.841
Rank: 49/113
0.53531
(±0.00061)
Rank: 35/113
521.37
Rank: 38/113
4473.48
Rank: 29/113
4.597
Rank: 15/113
0.414
Rank: 28/113
0.72179
(±0.00330)
Rank: 24/113
0.62855
Rank: 30/113
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
[sid:00076] SOSNet w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 575.53
Rank: 30/113
0.472
Rank: 77/113
0.846
Rank: 39/113
0.55170
(±0.00042)
Rank: 22/113
508.66
Rank: 43/113
4502.19
Rank: 26/113
4.377
Rank: 55/113
0.405
Rank: 23/113
0.71820
(±0.00298)
Rank: 29/113
0.63495
Rank: 24/113
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
[sid:00065] SOSNet-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 415.92
Rank: 64/113
0.486
Rank: 36/113
0.864
Rank: 21/113
0.54098
(±0.00086)
Rank: 29/113
521.37
Rank: 38/113
4473.48
Rank: 29/113
4.597
Rank: 15/113
0.417
Rank: 30/113
0.72179
(±0.00330)
Rank: 24/113
0.63139
Rank: 26/113
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
[sid:00522] SIFT and HardNet64 train sca...
Size: 512 bytes. Matches: built-in
7830.09 523.62
Rank: 39/113
0.486
Rank: 27/113
0.852
Rank: 34/113
0.55385
(±0.00076)
Rank: 20/113
822.07
Rank: 14/113
5660.19
Rank: 9/113
4.555
Rank: 24/113
0.404
Rank: 21/113
0.72744
(±0.00455)
Rank: 18/113
0.64065
Rank: 21/113
Anonymous (released after deadline) sift8k hardnet64-train-raw64 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-26 is_submission
[sid:00554] ContextDesc-Upright w/ DEGEN...
Size: 512 bytes. Matches: built-in
7830.09 480.10
Rank: 47/113
0.487
Rank: 23/113
0.877
Rank: 1/113
0.55697
(±0.00079)
Rank: 18/113
543.45
Rank: 29/113
4543.76
Rank: 25/113
4.498
Rank: 36/113
0.405
Rank: 22/113
0.72902
(±0.00255)
Rank: 17/113
0.64299
Rank: 20/113
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_baseline
[sid:00504] ASLFeat (MS) w/ DEGENSAC (no...
Size: 512 bytes. Matches: built-in
6948.83 396.95
Rank: 67/113
0.550
Rank: 9/113
0.797
Rank: 88/113
0.46102
(±0.00067)
Rank: 84/113
510.80
Rank: 40/113
3817.14
Rank: 62/113
4.577
Rank: 20/113
0.444
Rank: 63/113
0.68246
(±0.00173)
Rank: 63/113
0.57174
Rank: 76/113
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
[sid:00548] HarrisZ + DeepOrientation + ...
Size: 512 bytes. Matches: custom
2410.18 348.55
Rank: 84/113
0.368
Rank: 104/113
0.607
Rank: 105/113
0.28795
(±0.00000)
Rank: 101/113
351.53
Rank: 89/113
1699.08
Rank: 109/113
4.124
Rank: 94/113
0.507
Rank: 96/113
0.61370
(±0.00246)
Rank: 92/113
0.45082
Rank: 99/113
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
[sid:00519] SOSNet-Upright-AdaLAM
Size: 512 bytes. Matches: custom
7830.09 818.02
Rank: 10/113
0.486
Rank: 27/113
0.818
Rank: 80/113
0.56056
(±0.00000)
Rank: 16/113
827.85
Rank: 13/113
6157.62
Rank: 7/113
4.700
Rank: 3/113
0.381
Rank: 8/113
0.75917
(±0.00446)
Rank: 5/113
0.65987
Rank: 10/113
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. N/A N/A 20-04-26 is_submission
[sid:00101] KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 610.11
Rank: 23/113
0.582
Rank: 1/113
0.826
Rank: 71/113
0.49856
(±0.00038)
Rank: 64/113
356.21
Rank: 86/113
3366.01
Rank: 81/113
4.319
Rank: 64/113
0.464
Rank: 83/113
0.64829
(±0.00174)
Rank: 81/113
0.57342
Rank: 74/113
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
[sid:00501] Guided matching with Upright...
Size: 512 bytes. Matches: custom
7829.27 620.79
Rank: 19/113
0.486
Rank: 25/113
0.786
Rank: 90/113
0.49072
(±0.00000)
Rank: 68/113
1363.98
Rank: 2/113
7835.75
Rank: 1/113
4.219
Rank: 82/113
0.454
Rank: 74/113
0.68064
(±0.00390)
Rank: 66/113
0.58568
Rank: 66/113
Anonymous (released after deadline) sift upright-root-sift (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00060] L2-Net w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 396.66
Rank: 68/113
0.472
Rank: 77/113
0.813
Rank: 84/113
0.49054
(±0.00043)
Rank: 69/113
314.54
Rank: 105/113
3314.19
Rank: 87/113
4.204
Rank: 85/113
0.465
Rank: 84/113
0.65665
(±0.00287)
Rank: 75/113
0.57360
Rank: 73/113
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
[sid:00545]structured feature descriptio...
Size: 512 bytes. Matches: built-in
7830.21 420.34
Rank: 61/113
0.486
Rank: 34/113
0.865
Rank: 18/113
0.54442
(±0.00084)
Rank: 27/113
524.05
Rank: 36/113
4467.23
Rank: 31/113
4.551
Rank: 25/113
0.412
Rank: 25/113
0.72729
(±0.00300)
Rank: 19/113
0.63586
Rank: 23/113
Anonymous (released after deadline) sift sfd (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-29 is_submission
[sid:00120] CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 270.99
Rank: 98/113
0.472
Rank: 69/113
0.873
Rank: 9/113
0.46073
(±0.00062)
Rank: 85/113
356.56
Rank: 83/113
3550.63
Rank: 70/113
4.275
Rank: 72/113
0.439
Rank: 56/113
0.68875
(±0.00138)
Rank: 56/113
0.57474
Rank: 72/113
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
[sid:00568] Guided-SOSNet-lib-p
Size: 512 bytes. Matches: custom
7829.63 517.65
Rank: 41/113
0.486
Rank: 36/113
0.874
Rank: 8/113
0.57982
(±0.00000)
Rank: 10/113
524.66
Rank: 35/113
4618.86
Rank: 22/113
4.632
Rank: 8/113
0.369
Rank: 2/113
0.75888
(±0.00437)
Rank: 6/113
0.66935
Rank: 4/113
Anonymous (released after deadline) sift8k sosnet-lib-p (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-15 is_baseline
[sid:00051] HardNet w/ DEGENSAC (no FLAN...
Size: 512 bytes. Matches: built-in
7861.11 441.30
Rank: 58/113
0.472
Rank: 77/113
0.866
Rank: 17/113
0.55430
(±0.00031)
Rank: 19/113
426.84
Rank: 61/113
4001.40
Rank: 54/113
4.339
Rank: 62/113
0.419
Rank: 34/113
0.70962
(±0.00191)
Rank: 36/113
0.63196
Rank: 25/113
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
[sid:00518] UprightRootSIFT-AdaLAM
Size: 512 bytes. Matches: custom
6449.42 443.61
Rank: 57/113
0.436
Rank: 101/113
0.825
Rank: 73/113
0.49625
(±0.00000)
Rank: 65/113
449.49
Rank: 56/113
4353.56
Rank: 37/113
4.453
Rank: 42/113
0.397
Rank: 15/113
0.72490
(±0.00190)
Rank: 20/113
0.61057
Rank: 46/113
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. N/A N/A 20-04-24 is_submission
[sid:00058] HardNet-Upright w/ MAGSAC (n...
Size: 512 bytes. Matches: built-in
7829.63 722.10
Rank: 13/113
0.486
Rank: 36/113
0.853
Rank: 27/113
0.56374
(±0.00069)
Rank: 15/113
509.07
Rank: 41/113
4250.40
Rank: 38/113
4.548
Rank: 26/113
0.412
Rank: 26/113
0.72309
(±0.00141)
Rank: 22/113
0.64341
Rank: 19/113
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
[sid:00122] CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 539.84
Rank: 34/113
0.472
Rank: 69/113
0.838
Rank: 51/113
0.53291
(±0.00047)
Rank: 37/113
356.56
Rank: 83/113
3550.63
Rank: 70/113
4.275
Rank: 72/113
0.443
Rank: 62/113
0.68875
(±0.00138)
Rank: 56/113
0.61083
Rank: 45/113
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
[sid:00061] L2-Net-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 366.70
Rank: 78/113
0.486
Rank: 36/113
0.845
Rank: 40/113
0.52012
(±0.00106)
Rank: 49/113
369.60
Rank: 81/113
3538.91
Rank: 73/113
4.407
Rank: 45/113
0.447
Rank: 68/113
0.68108
(±0.00177)
Rank: 64/113
0.60060
Rank: 56/113
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
[sid:00121] CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 406.63
Rank: 66/113
0.472
Rank: 69/113
0.863
Rank: 22/113
0.53852
(±0.00039)
Rank: 32/113
356.56
Rank: 83/113
3550.63
Rank: 70/113
4.275
Rank: 72/113
0.439
Rank: 55/113
0.68875
(±0.00138)
Rank: 56/113
0.61364
Rank: 39/113
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
[sid:00069] GeoDesc w/ DEGENSAC (no FLAN...
Size: 512 bytes. Matches: built-in
7861.11 356.12
Rank: 83/113
0.472
Rank: 77/113
0.856
Rank: 24/113
0.51112
(±0.00070)
Rank: 57/113
395.12
Rank: 76/113
3838.97
Rank: 60/113
4.264
Rank: 75/113
0.443
Rank: 60/113
0.68032
(±0.00099)
Rank: 67/113
0.59572
Rank: 62/113
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
[sid:00591] Guided-Hardnet-Epoch2
Size: 512 bytes. Matches: custom
7829.63 774.81
Rank: 11/113
0.486
Rank: 36/113
0.823
Rank: 77/113
0.61081
(±0.00000)
Rank: 1/113
785.57
Rank: 17/113
6330.70
Rank: 4/113
4.680
Rank: 4/113
0.358
Rank: 1/113
0.78288
(±0.00094)
Rank: 1/113
0.69684
Rank: 1/113
Anonymous (released after deadline) sift8k hardnet-epoch2 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-24 is_baseline
[sid:00086] LogPolar-Upright w/ MAGSAC (...
Size: 512 bytes. Matches: built-in
7829.63 746.24
Rank: 12/113
0.486
Rank: 36/113
0.844
Rank: 45/113
0.54060
(±0.00047)
Rank: 30/113
505.37
Rank: 46/113
4414.11
Rank: 32/113
4.518
Rank: 33/113
0.421
Rank: 38/113
0.71092
(±0.00251)
Rank: 33/113
0.62576
Rank: 33/113
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
[sid:00106] VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 333.93
Rank: 88/113
0.490
Rank: 18/113
0.809
Rank: 85/113
0.46326
(±0.00089)
Rank: 82/113
324.62
Rank: 102/113
3030.67
Rank: 96/113
4.173
Rank: 87/113
0.462
Rank: 81/113
0.62829
(±0.00112)
Rank: 89/113
0.54578
Rank: 84/113
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
[sid:00081] LogPolar w/ MAGSAC (no FLANN...
Size: 512 bytes. Matches: built-in
7861.11 603.70
Rank: 24/113
0.472
Rank: 77/113
0.832
Rank: 61/113
0.52385
(±0.00113)
Rank: 44/113
415.77
Rank: 66/113
4054.60
Rank: 49/113
4.316
Rank: 67/113
0.432
Rank: 51/113
0.69284
(±0.00429)
Rank: 51/113
0.60835
Rank: 50/113
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
[sid:00562] HardNet64-train-all-raw64-ba...
Size: 512 bytes. Matches: built-in
7831.92 377.01
Rank: 74/113
0.486
Rank: 64/113
0.807
Rank: 86/113
0.45770
(±0.00060)
Rank: 86/113
575.43
Rank: 26/113
4756.46
Rank: 18/113
4.563
Rank: 23/113
0.435
Rank: 54/113
0.69780
(±0.00191)
Rank: 47/113
0.57775
Rank: 70/113
Anonymous (released after deadline) sift8k hardnet64-train-all-raw64-balance-l2-224000-raw64 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-13 is_baseline
[sid:00557] HarrisZ + DeepOrientation + ...
Size: 512 bytes. Matches: custom
2410.18 622.70
Rank: 18/113
0.368
Rank: 104/113
0.513
Rank: 107/113
0.05482
(±0.00000)
Rank: 107/113
626.57
Rank: 20/113
1793.00
Rank: 108/113
4.314
Rank: 69/113
0.517
Rank: 99/113
0.56933
(±0.00268)
Rank: 102/113
0.31207
Rank: 104/113
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_baseline
[sid:00520] SIFT and HardNet64 train sca...
Size: 512 bytes. Matches: built-in
7830.09 408.60
Rank: 65/113
0.486
Rank: 27/113
0.837
Rank: 56/113
0.50589
(±0.00059)
Rank: 59/113
619.25
Rank: 21/113
4959.78
Rank: 14/113
4.384
Rank: 48/113
0.456
Rank: 76/113
0.69172
(±0.00048)
Rank: 53/113
0.59880
Rank: 58/113
Anonymous (released after deadline) sift8k hardnet64-train-scale5 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00526] ASLFeat-MS
Size: 512 bytes. Matches: built-in
7384.19 464.39
Rank: 50/113
0.578
Rank: 4/113
0.752
Rank: 98/113
0.42208
(±0.00054)
Rank: 90/113
532.83
Rank: 33/113
4002.80
Rank: 53/113
4.521
Rank: 32/113
0.447
Rank: 65/113
0.66568
(±0.00347)
Rank: 72/113
0.54388
Rank: 85/113
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. Under preparation N/A 20-04-27 is_submission
[sid:00546] HarrisZ + DeepOrientation + ...
Size: 512 bytes. Matches: custom
2410.18 1036.77
Rank: 5/113
0.368
Rank: 104/113
0.366
Rank: 110/113
0.07780
(±0.00000)
Rank: 106/113
1047.01
Rank: 6/113
2197.09
Rank: 105/113
4.023
Rank: 101/113
0.509
Rank: 97/113
0.59222
(±0.00276)
Rank: 97/113
0.33501
Rank: 101/113
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
[sid:00524] SIFT-patch
Size: 512 bytes. Matches: built-in
7861.62 85.68
Rank: 113/113
0.472
Rank: 74/113
0.574
Rank: 106/113
0.22805
(±0.00076)
Rank: 103/113
81.93
Rank: 113/113
991.42
Rank: 113/113
1.934
Rank: 113/113
0.687
Rank: 110/113
0.08662
(±0.00384)
Rank: 112/113
0.15733
Rank: 112/113
feyman_priv (contact) sift8k l2net-arcface (128 float32: 512 bytes) sift8k with l2net trained on google-landmark-dataset-v1(in 1000 class) N/A 20-04-26 is_submission
[sid:00082] LogPolar w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 507.35
Rank: 42/113
0.472
Rank: 77/113
0.817
Rank: 81/113
0.49129
(±0.00011)
Rank: 67/113
399.70
Rank: 71/113
4036.37
Rank: 51/113
4.340
Rank: 60/113
0.429
Rank: 49/113
0.69100
(±0.00154)
Rank: 54/113
0.59115
Rank: 65/113
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
[sid:00116] VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 214.47
Rank: 105/113
0.577
Rank: 5/113
0.841
Rank: 50/113
0.39330
(±0.00052)
Rank: 94/113
350.69
Rank: 90/113
3327.71
Rank: 84/113
4.076
Rank: 98/113
0.490
Rank: 94/113
0.60691
(±0.00338)
Rank: 93/113
0.50011
Rank: 96/113
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
[sid:00063] SOSNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 347.73
Rank: 85/113
0.472
Rank: 77/113
0.854
Rank: 26/113
0.51891
(±0.00067)
Rank: 50/113
440.21
Rank: 58/113
4178.82
Rank: 41/113
4.378
Rank: 53/113
0.419
Rank: 35/113
0.70078
(±0.00375)
Rank: 45/113
0.60984
Rank: 48/113
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
[sid:00109] VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 255.82
Rank: 100/113
0.482
Rank: 66/113
0.838
Rank: 53/113
0.46795
(±0.00075)
Rank: 78/113
311.54
Rank: 107/113
3061.54
Rank: 93/113
4.105
Rank: 95/113
0.475
Rank: 87/113
0.62964
(±0.00428)
Rank: 85/113
0.54880
Rank: 81/113
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
[sid:00118] VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 358.98
Rank: 81/113
0.577
Rank: 5/113
0.824
Rank: 75/113
0.46263
(±0.00079)
Rank: 83/113
350.69
Rank: 90/113
3327.71
Rank: 84/113
4.076
Rank: 98/113
0.489
Rank: 93/113
0.60691
(±0.00338)
Rank: 93/113
0.53477
Rank: 87/113
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
[sid:00576] Upright-Sift + X-Net-lib w/ ...
Size: 512 bytes. Matches: built-in
7830.16 475.12
Rank: 48/113
0.486
Rank: 35/113
0.844
Rank: 43/113
0.53156
(±0.00027)
Rank: 38/113
720.29
Rank: 18/113
5398.84
Rank: 11/113
4.547
Rank: 28/113
0.408
Rank: 24/113
0.71966
(±0.00160)
Rank: 28/113
0.62561
Rank: 34/113
Anonymous (released after deadline) sift x-net-lib (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-20 is_baseline
[sid:00075] SOSNet w/ DEGENSAC (no FLANN...
Size: 512 bytes. Matches: built-in
7861.11 433.37
Rank: 60/113
0.472
Rank: 77/113
0.868
Rank: 14/113
0.55867
(±0.00060)
Rank: 17/113
508.66
Rank: 43/113
4502.19
Rank: 26/113
4.377
Rank: 55/113
0.404
Rank: 20/113
0.71820
(±0.00298)
Rank: 29/113
0.63844
Rank: 22/113
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
[sid:00105] VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 267.54
Rank: 99/113
0.490
Rank: 18/113
0.826
Rank: 72/113
0.46555
(±0.00055)
Rank: 81/113
324.62
Rank: 102/113
3030.67
Rank: 96/113
4.173
Rank: 87/113
0.461
Rank: 79/113
0.62829
(±0.00112)
Rank: 89/113
0.54692
Rank: 83/113
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
[sid:00062] L2-Net-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 483.13
Rank: 46/113
0.486
Rank: 36/113
0.822
Rank: 78/113
0.51450
(±0.00058)
Rank: 53/113
369.60
Rank: 81/113
3538.91
Rank: 73/113
4.407
Rank: 45/113
0.448
Rank: 70/113
0.68108
(±0.00177)
Rank: 64/113
0.59779
Rank: 59/113
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
[sid:00048] GeoDesc w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 371.02
Rank: 77/113
0.472
Rank: 77/113
0.816
Rank: 82/113
0.46887
(±0.00025)
Rank: 77/113
350.43
Rank: 93/113
3601.67
Rank: 67/113
4.234
Rank: 79/113
0.447
Rank: 67/113
0.65553
(±0.00444)
Rank: 78/113
0.56220
Rank: 79/113
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
[sid:00056] HardNet-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 448.53
Rank: 55/113
0.486
Rank: 36/113
0.864
Rank: 20/113
0.54393
(±0.00027)
Rank: 28/113
480.13
Rank: 50/113
4167.34
Rank: 43/113
4.566
Rank: 21/113
0.417
Rank: 31/113
0.71542
(±0.00274)
Rank: 31/113
0.62967
Rank: 29/113
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
[sid:00049] GeoDesc-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 324.65
Rank: 91/113
0.486
Rank: 36/113
0.850
Rank: 35/113
0.48769
(±0.00064)
Rank: 71/113
394.67
Rank: 78/113
3863.42
Rank: 58/113
4.380
Rank: 51/113
0.440
Rank: 57/113
0.67682
(±0.00127)
Rank: 69/113
0.58225
Rank: 67/113
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
[sid:00057] HardNet-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 595.31
Rank: 26/113
0.486
Rank: 36/113
0.841
Rank: 48/113
0.53730
(±0.00082)
Rank: 34/113
480.13
Rank: 50/113
4167.34
Rank: 43/113
4.566
Rank: 21/113
0.416
Rank: 29/113
0.71542
(±0.00274)
Rank: 31/113
0.62636
Rank: 32/113
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
[sid:00047] GeoDesc w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 287.87
Rank: 95/113
0.472
Rank: 77/113
0.838
Rank: 54/113
0.47259
(±0.00033)
Rank: 76/113
350.43
Rank: 93/113
3601.67
Rank: 67/113
4.234
Rank: 79/113
0.447
Rank: 66/113
0.65553
(±0.00444)
Rank: 78/113
0.56406
Rank: 78/113
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
[sid:00558] SIFT + DeepOrientation + SOS...
Size: 512 bytes. Matches: custom
2826.82 339.17
Rank: 86/113
0.284
Rank: 109/113
0.427
Rank: 109/113
0.01806
(±0.00000)
Rank: 111/113
341.75
Rank: 98/113
2003.56
Rank: 106/113
3.735
Rank: 105/113
0.593
Rank: 104/113
0.51048
(±0.00436)
Rank: 104/113
0.26427
Rank: 108/113
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_baseline
[sid:00079] LogPolar w/ DEGENSAC (no FLA...
Size: 512 bytes. Matches: built-in
7861.11 450.82
Rank: 54/113
0.472
Rank: 77/113
0.852
Rank: 29/113
0.53396
(±0.00086)
Rank: 36/113
415.77
Rank: 66/113
4054.60
Rank: 49/113
4.316
Rank: 67/113
0.432
Rank: 53/113
0.69284
(±0.00429)
Rank: 51/113
0.61340
Rank: 40/113
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
[sid:00077] SOSNet-Upright w/ DEGENSAC (...
Size: 512 bytes. Matches: built-in
7829.63 518.08
Rank: 40/113
0.486
Rank: 36/113
0.877
Rank: 2/113
0.57385
(±0.00041)
Rank: 11/113
600.23
Rank: 23/113
4765.43
Rank: 16/113
4.584
Rank: 18/113
0.403
Rank: 18/113
0.73027
(±0.00284)
Rank: 15/113
0.65206
Rank: 14/113
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
[sid:00502] Guided matching with Upright...
Size: 512 bytes. Matches: custom
7829.27 501.50
Rank: 43/113
0.486
Rank: 25/113
0.828
Rank: 69/113
0.50310
(±0.01761)
Rank: 61/113
809.76
Rank: 15/113
5966.91
Rank: 8/113
4.462
Rank: 40/113
0.425
Rank: 45/113
0.71009
(±0.00414)
Rank: 35/113
0.60659
Rank: 53/113
Anonymous (released after deadline) sift upright-root-sift (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00015] Root-SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
7860.73 281.05
Rank: 97/113
0.472
Rank: 72/113
0.845
Rank: 41/113
0.48887
(±0.00011)
Rank: 70/113
437.74
Rank: 60/113
3814.80
Rank: 63/113
4.151
Rank: 90/113
0.458
Rank: 77/113
0.65067
(±0.00155)
Rank: 80/113
0.56977
Rank: 77/113
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
[sid:00542] sift and hardnet64 train sca...
Size: 512 bytes. Matches: built-in
7830.09 618.78
Rank: 20/113
0.486
Rank: 27/113
0.872
Rank: 10/113
0.58801
(±0.00037)
Rank: 3/113
950.74
Rank: 9/113
6286.36
Rank: 6/113
4.626
Rank: 9/113
0.390
Rank: 14/113
0.74662
(±0.00087)
Rank: 11/113
0.66732
Rank: 5/113
Anonymous (released after deadline) sift8k hardnet64-train-all-l2 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-01 is_submission
[sid:00535] hardnetlib32P-Upright w/ DEG...
Size: 512 bytes. Matches: built-in
7829.63 537.62
Rank: 35/113
0.486
Rank: 36/113
0.876
Rank: 4/113
0.57327
(±0.00078)
Rank: 12/113
506.73
Rank: 45/113
4230.98
Rank: 40/113
4.517
Rank: 35/113
0.418
Rank: 33/113
0.72171
(±0.00156)
Rank: 26/113
0.64749
Rank: 18/113
Anonymous (released after deadline) sift8k-no-dups-mps hardnetlib32P (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-27 is_submission
[sid:00114] VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 358.21
Rank: 82/113
0.547
Rank: 11/113
0.800
Rank: 87/113
0.43347
(±0.00084)
Rank: 89/113
347.39
Rank: 95/113
3209.10
Rank: 89/113
4.126
Rank: 91/113
0.517
Rank: 98/113
0.58657
(±0.00120)
Rank: 99/113
0.51002
Rank: 95/113
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
[sid:00563] HardNet64-train-all-SOS-8120...
Size: 512 bytes. Matches: built-in
7830.09 624.02
Rank: 16/113
0.486
Rank: 27/113
0.869
Rank: 13/113
0.58664
(±0.00095)
Rank: 4/113
972.47
Rank: 7/113
6369.03
Rank: 3/113
4.612
Rank: 11/113
0.388
Rank: 12/113
0.74668
(±0.00251)
Rank: 10/113
0.66666
Rank: 6/113
Anonymous (released after deadline) sift8k hardnet64-train-all-sos-812000 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-13 is_baseline
[sid:00055] HardNet-Upright w/ DEGENSAC ...
Size: 512 bytes. Matches: built-in
7829.63 537.62
Rank: 36/113
0.486
Rank: 36/113
0.876
Rank: 3/113
0.57279
(±0.00120)
Rank: 13/113
509.07
Rank: 41/113
4250.40
Rank: 38/113
4.548
Rank: 26/113
0.413
Rank: 27/113
0.72309
(±0.00141)
Rank: 22/113
0.64794
Rank: 17/113
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
[sid:00108] VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 175.46
Rank: 108/113
0.482
Rank: 66/113
0.848
Rank: 37/113
0.39839
(±0.00122)
Rank: 93/113
311.54
Rank: 107/113
3061.54
Rank: 93/113
4.105
Rank: 95/113
0.475
Rank: 89/113
0.62964
(±0.00428)
Rank: 85/113
0.51401
Rank: 93/113
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
CV-DoG-HardNet8-PT
Size: 512 bytes. Matches: built-in
7829.26 589.94
Rank: 28/113
0.484
Rank: 65/113
0.872
Rank: 11/113
0.58375
(±0.00046)
Rank: 6/113
573.22
Rank: 27/113
4489.23
Rank: 28/113
4.619
Rank: 10/113
0.403
Rank: 19/113
0.72432
(±0.00357)
Rank: 21/113
0.65403
Rank: 12/113
Milan Pultar, Dmytro Mishkin, Jiri Matas (contact) sift8k h8e512pt (128 float32: 512 bytes) [sid:00594] HardNet8 with PCA compression N/A N/A 20-05-27 is_baseline
[sid:00533] upright-sift8k-hardnet
Size: 512 bytes. Matches: built-in
6589.88 379.73
Rank: 73/113
0.467
Rank: 97/113
0.852
Rank: 32/113
0.52463
(±0.00070)
Rank: 43/113
404.02
Rank: 68/113
3501.62
Rank: 76/113
4.483
Rank: 38/113
0.428
Rank: 47/113
0.70123
(±0.00081)
Rank: 44/113
0.61293
Rank: 41/113
Anonymous (released after deadline) sift8k hardnet (128 float32: 512 bytes) Anonymous (released after deadline) https://arxiv.org/abs/1705.10872 Anonymous (released after deadline) 20-04-24 is_submission
[sid:00053] HardNet w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 587.65
Rank: 29/113
0.472
Rank: 77/113
0.842
Rank: 47/113
0.55022
(±0.00001)
Rank: 24/113
426.84
Rank: 61/113
4001.40
Rank: 54/113
4.339
Rank: 62/113
0.419
Rank: 37/113
0.70962
(±0.00191)
Rank: 36/113
0.62992
Rank: 28/113
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
[sid:00005] SURF (OpenCV), DEGENSAC
Size: 256 bytes. Matches: built-in
7728.57 128.86
Rank: 112/113
0.432
Rank: 103/113
0.684
Rank: 100/113
0.24948
(±0.00065)
Rank: 102/113
928.29
Rank: 10/113
3455.28
Rank: 77/113
3.428
Rank: 108/113
0.733
Rank: 111/113
0.40490
(±0.00448)
Rank: 110/113
0.32719
Rank: 103/113
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
[sid:00510] sift and hardnet64
Size: 512 bytes. Matches: built-in
7861.62 333.97
Rank: 87/113
0.472
Rank: 75/113
0.845
Rank: 42/113
0.51149
(±0.00018)
Rank: 56/113
521.58
Rank: 37/113
4587.88
Rank: 23/113
4.222
Rank: 81/113
0.424
Rank: 44/113
0.69458
(±0.00340)
Rank: 50/113
0.60304
Rank: 54/113
Anonymous (released after deadline) sift8k hardnet64-v2 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00559] SIFT + DeepOrientation + SOS...
Size: 512 bytes. Matches: custom
2826.82 166.47
Rank: 109/113
0.284
Rank: 109/113
0.650
Rank: 102/113
0.05206
(±0.00000)
Rank: 108/113
167.59
Rank: 112/113
1549.91
Rank: 110/113
3.989
Rank: 104/113
0.622
Rank: 106/113
0.47700
(±0.00167)
Rank: 106/113
0.26453
Rank: 107/113
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_baseline
[sid:00070] GeoDesc-Upright w/ MAGSAC (n...
Size: 512 bytes. Matches: built-in
7829.63 545.49
Rank: 33/113
0.486
Rank: 36/113
0.848
Rank: 38/113
0.52078
(±0.00028)
Rank: 48/113
458.60
Rank: 53/113
4146.83
Rank: 46/113
4.412
Rank: 43/113
0.432
Rank: 52/113
0.70442
(±0.00189)
Rank: 42/113
0.61260
Rank: 43/113
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
[sid:00052] HardNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 371.67
Rank: 76/113
0.472
Rank: 77/113
0.853
Rank: 28/113
0.52325
(±0.00098)
Rank: 45/113
402.15
Rank: 69/113
3895.00
Rank: 56/113
4.345
Rank: 57/113
0.426
Rank: 46/113
0.69653
(±0.00108)
Rank: 48/113
0.60989
Rank: 47/113
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
[sid:00549] SIFT + DeepOrientation + SOS...
Size: 512 bytes. Matches: custom
2826.82 219.38
Rank: 104/113
0.284
Rank: 109/113
0.456
Rank: 108/113
0.14544
(±0.00000)
Rank: 105/113
221.40
Rank: 111/113
1839.87
Rank: 107/113
3.608
Rank: 106/113
0.641
Rank: 107/113
0.47073
(±0.00158)
Rank: 107/113
0.30808
Rank: 105/113
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
[sid:00113] VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 297.80
Rank: 94/113
0.547
Rank: 11/113
0.814
Rank: 83/113
0.44501
(±0.00077)
Rank: 88/113
347.39
Rank: 95/113
3209.10
Rank: 89/113
4.126
Rank: 91/113
0.519
Rank: 101/113
0.58657
(±0.00120)
Rank: 99/113
0.51579
Rank: 91/113
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
[sid:00567] Guided-HardNet32-v1-lib-qht-...
Size: 512 bytes. Matches: custom
7829.63 529.89
Rank: 37/113
0.486
Rank: 36/113
0.875
Rank: 6/113
0.58509
(±0.00000)
Rank: 5/113
536.93
Rank: 32/113
4685.16
Rank: 21/113
4.645
Rank: 6/113
0.376
Rank: 5/113
0.75702
(±0.00159)
Rank: 8/113
0.67105
Rank: 3/113
Anonymous (released after deadline) sift8k hardnet32-v1-lib-qht-p (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-15 is_baseline
[sid:00547] SIFT + DeepOrientation + SOS...
Size: 512 bytes. Matches: custom
2826.82 1096.89
Rank: 3/113
0.284
Rank: 109/113
0.166
Rank: 112/113
0.04936
(±0.00000)
Rank: 109/113
1106.68
Rank: 4/113
2494.00
Rank: 103/113
3.400
Rank: 109/113
0.646
Rank: 108/113
0.43054
(±0.00185)
Rank: 108/113
0.23995
Rank: 110/113
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
[sid:00537] SIFT + DeepOrientation + SOS...
Size: 512 bytes. Matches: custom
2826.82 1496.97
Rank: 1/113
0.284
Rank: 109/113
0.145
Rank: 113/113
0.00178
(±0.00000)
Rank: 113/113
1506.82
Rank: 1/113
2538.44
Rank: 102/113
3.388
Rank: 111/113
0.669
Rank: 109/113
0.42682
(±0.00259)
Rank: 109/113
0.21430
Rank: 111/113
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
[sid:00110] VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 324.67
Rank: 90/113
0.482
Rank: 66/113
0.820
Rank: 79/113
0.46657
(±0.00079)
Rank: 80/113
311.54
Rank: 107/113
3061.54
Rank: 93/113
4.105
Rank: 95/113
0.475
Rank: 88/113
0.62964
(±0.00428)
Rank: 85/113
0.54810
Rank: 82/113
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
[sid:00072] L2-Net-Upright w/ DEGENSAC (...
Size: 512 bytes. Matches: built-in
7829.63 444.75
Rank: 56/113
0.486
Rank: 36/113
0.859
Rank: 23/113
0.54497
(±0.00003)
Rank: 26/113
395.53
Rank: 74/113
3603.85
Rank: 65/113
4.382
Rank: 49/113
0.452
Rank: 73/113
0.68491
(±0.00338)
Rank: 61/113
0.61494
Rank: 38/113
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
[sid:00590] Guided-HardNet-epoch4
Size: 512 bytes. Matches: custom
7829.63 596.85
Rank: 25/113
0.486
Rank: 36/113
0.875
Rank: 5/113
0.59919
(±0.00000)
Rank: 2/113
604.81
Rank: 22/113
5062.27
Rank: 13/113
4.710
Rank: 2/113
0.369
Rank: 3/113
0.76219
(±0.00253)
Rank: 3/113
0.68069
Rank: 2/113
Anonymous (released after deadline) sift8k hardnet-epoch4 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-24 is_baseline
[sid:00117] VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 306.56
Rank: 92/113
0.577
Rank: 5/113
0.831
Rank: 62/113
0.46793
(±0.00035)
Rank: 79/113
350.69
Rank: 90/113
3327.71
Rank: 84/113
4.076
Rank: 98/113
0.487
Rank: 91/113
0.60691
(±0.00338)
Rank: 93/113
0.53742
Rank: 86/113
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
[sid:00074] L2-Net w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 492.41
Rank: 44/113
0.472
Rank: 77/113
0.830
Rank: 68/113
0.52519
(±0.00039)
Rank: 41/113
339.00
Rank: 99/113
3424.94
Rank: 78/113
4.206
Rank: 83/113
0.448
Rank: 69/113
0.66445
(±0.00213)
Rank: 73/113
0.59482
Rank: 63/113
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
[sid:00595] SIFT8k_8000_HardNet64-train-...
Size: 512 bytes. Matches: built-in
7830.09 611.66
Rank: 22/113
0.486
Rank: 27/113
0.867
Rank: 15/113
0.58265
(±0.00102)
Rank: 8/113
786.40
Rank: 16/113
5595.19
Rank: 10/113
4.603
Rank: 14/113
0.390
Rank: 13/113
0.74372
(±0.00317)
Rank: 14/113
0.66319
Rank: 9/113
Anonymous (released after deadline) sift8k hardnet64-train-all-ben86-l2-210000 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-27 is_baseline
[sid:00521] SIFT8k-gift
Size: 512 bytes. Matches: built-in
6589.48 435.27
Rank: 59/113
0.467
Rank: 99/113
0.830
Rank: 64/113
0.48128
(±0.00056)
Rank: 74/113
452.18
Rank: 55/113
3770.17
Rank: 64/113
4.595
Rank: 17/113
0.444
Rank: 64/113
0.67443
(±0.00253)
Rank: 71/113
0.57785
Rank: 69/113
Anonymous (released after deadline) sift-def gift (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-26 is_submission
[sid:00566] Guided-HardNet2qhtp.json
Size: 512 bytes. Matches: custom
7829.63 527.15
Rank: 38/113
0.486
Rank: 36/113
0.874
Rank: 7/113
0.58251
(±0.00000)
Rank: 9/113
448.12
Rank: 57/113
4072.82
Rank: 48/113
4.607
Rank: 12/113
0.387
Rank: 11/113
0.74640
(±0.00246)
Rank: 12/113
0.66445
Rank: 7/113
Anonymous (released after deadline) sift8k hardnetnd32qhtp (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-15 is_baseline
[sid:00572] Guided HardNet qht
Size: 512 bytes. Matches: custom
7829.63 899.89
Rank: 7/113
0.486
Rank: 36/113
0.774
Rank: 94/113
0.54663
(±0.00000)
Rank: 25/113
538.28
Rank: 30/113
4687.01
Rank: 20/113
4.644
Rank: 7/113
0.376
Rank: 6/113
0.75721
(±0.00140)
Rank: 7/113
0.65192
Rank: 15/113
Anonymous (released after deadline) sift8k hardnet-qht (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-20 is_baseline
[sid:00541] sift and sosnet64 train scal...
Size: 512 bytes. Matches: built-in
7830.09 615.09
Rank: 21/113
0.486
Rank: 27/113
0.866
Rank: 16/113
0.58268
(±0.00061)
Rank: 7/113
957.95
Rank: 8/113
6312.38
Rank: 5/113
4.607
Rank: 13/113
0.385
Rank: 9/113
0.74476
(±0.00137)
Rank: 13/113
0.66372
Rank: 8/113
Anonymous (released after deadline) sift8k hardnet64-train-all-sos (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-01 is_submission
[sid:00102] KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 832.10
Rank: 9/113
0.582
Rank: 1/113
0.788
Rank: 89/113
0.47395
(±0.00073)
Rank: 75/113
356.21
Rank: 86/113
3366.01
Rank: 81/113
4.319
Rank: 64/113
0.464
Rank: 82/113
0.64829
(±0.00174)
Rank: 81/113
0.56112
Rank: 80/113
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
[sid:00556] HarrisZ + DeepOrientation + ...
Size: 512 bytes. Matches: custom
2410.18 390.12
Rank: 69/113
0.368
Rank: 104/113
0.646
Rank: 104/113
0.04192
(±0.00000)
Rank: 110/113
392.52
Rank: 80/113
1475.19
Rank: 111/113
4.454
Rank: 41/113
0.601
Rank: 105/113
0.48498
(±0.00118)
Rank: 105/113
0.26345
Rank: 109/113
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_baseline
[sid:00073] L2-Net-Upright w/ MAGSAC (no...
Size: 512 bytes. Matches: built-in
7829.63 590.66
Rank: 27/113
0.486
Rank: 36/113
0.837
Rank: 57/113
0.53911
(±0.00062)
Rank: 31/113
395.53
Rank: 74/113
3603.85
Rank: 65/113
4.382
Rank: 49/113
0.455
Rank: 75/113
0.68491
(±0.00338)
Rank: 61/113
0.61201
Rank: 44/113
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
[sid:00080] LogPolar w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 384.53
Rank: 72/113
0.472
Rank: 77/113
0.837
Rank: 55/113
0.50188
(±0.00044)
Rank: 62/113
399.70
Rank: 71/113
4036.37
Rank: 51/113
4.340
Rank: 60/113
0.429
Rank: 48/113
0.69100
(±0.00154)
Rank: 54/113
0.59644
Rank: 61/113
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
[sid:00006] Upright SIFT (OpenCV), DEGEN...
Size: 512 bytes. Matches: built-in
7829.24 326.21
Rank: 89/113
0.487
Rank: 21/113
0.830
Rank: 66/113
0.48742
(±0.00042)
Rank: 72/113
525.57
Rank: 34/113
4147.20
Rank: 45/113
4.258
Rank: 77/113
0.460
Rank: 78/113
0.65664
(±0.00231)
Rank: 77/113
0.57203
Rank: 75/113
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
[sid:00059] L2-Net w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 304.61
Rank: 93/113
0.472
Rank: 77/113
0.835
Rank: 60/113
0.49523
(±0.00042)
Rank: 66/113
314.54
Rank: 105/113
3314.19
Rank: 87/113
4.204
Rank: 85/113
0.467
Rank: 85/113
0.65665
(±0.00287)
Rank: 75/113
0.57594
Rank: 71/113
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
[sid:00050] GeoDesc-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 419.68
Rank: 62/113
0.486
Rank: 36/113
0.827
Rank: 70/113
0.48259
(±0.00060)
Rank: 73/113
394.67
Rank: 78/113
3863.42
Rank: 58/113
4.380
Rank: 51/113
0.443
Rank: 61/113
0.67682
(±0.00127)
Rank: 69/113
0.57971
Rank: 68/113
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
[sid:00054] HardNet w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 488.84
Rank: 45/113
0.472
Rank: 77/113
0.830
Rank: 67/113
0.51809
(±0.00053)
Rank: 51/113
402.15
Rank: 69/113
3895.00
Rank: 56/113
4.345
Rank: 57/113
0.423
Rank: 43/113
0.69653
(±0.00108)
Rank: 48/113
0.60731
Rank: 52/113
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
[sid:00067] GeoDesc w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 463.81
Rank: 51/113
0.472
Rank: 77/113
0.835
Rank: 58/113
0.50559
(±0.00081)
Rank: 60/113
395.12
Rank: 76/113
3838.97
Rank: 60/113
4.264
Rank: 75/113
0.442
Rank: 59/113
0.68032
(±0.00099)
Rank: 67/113
0.59295
Rank: 64/113
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
[sid:00540] ASLV2+OANetV2+DEGENSAC
Size: 512 bytes. Matches: custom
5982.87 1053.74
Rank: 4/113
0.550
Rank: 10/113
0.758
Rank: 96/113
0.50173
(±0.00000)
Rank: 63/113
1067.11
Rank: 5/113
4587.47
Rank: 24/113
4.948
Rank: 1/113
0.386
Rank: 10/113
0.75275
(±0.00186)
Rank: 9/113
0.62724
Rank: 31/113
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
[sid:00514] UprightRootSIFT-AdaLAM
Size: 512 bytes. Matches: custom
6449.42 632.63
Rank: 15/113
0.436
Rank: 101/113
0.758
Rank: 97/113
0.31372
(±0.00000)
Rank: 99/113
640.08
Rank: 19/113
5286.27
Rank: 12/113
4.343
Rank: 59/113
0.398
Rank: 16/113
0.72130
(±0.00085)
Rank: 27/113
0.51751
Rank: 90/113
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. N/A N/A 20-04-25 is_submission
[sid:00511] r2d2-wasfi
Size: 512 bytes. Matches: built-in
7861.03 281.30
Rank: 96/113
0.503
Rank: 16/113
0.762
Rank: 95/113
0.37893
(±0.00023)
Rank: 95/113
416.35
Rank: 65/113
2834.69
Rank: 99/113
4.293
Rank: 71/113
0.480
Rank: 90/113
0.64359
(±0.00103)
Rank: 84/113
0.51126
Rank: 94/113
Anonymous (released after deadline) r2d2 r2d2-wasfi-epoch25-pretrained-n16-8k (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00083] LogPolar-Upright w/ DEGENSAC...
Size: 512 bytes. Matches: built-in
7829.63 553.21
Rank: 31/113
0.486
Rank: 36/113
0.865
Rank: 19/113
0.55102
(±0.00015)
Rank: 23/113
505.37
Rank: 46/113
4414.11
Rank: 32/113
4.518
Rank: 33/113
0.422
Rank: 40/113
0.71092
(±0.00251)
Rank: 33/113
0.63097
Rank: 27/113
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
[sid:00064] SOSNet w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 455.02
Rank: 53/113
0.472
Rank: 77/113
0.831
Rank: 63/113
0.51402
(±0.00057)
Rank: 54/113
440.21
Rank: 58/113
4178.82
Rank: 41/113
4.378
Rank: 53/113
0.422
Rank: 41/113
0.70078
(±0.00375)
Rank: 45/113
0.60740
Rank: 51/113
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
[sid:00104] VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 183.65
Rank: 107/113
0.490
Rank: 18/113
0.835
Rank: 59/113
0.39987
(±0.00068)
Rank: 91/113
324.62
Rank: 102/113
3030.67
Rank: 96/113
4.173
Rank: 87/113
0.467
Rank: 86/113
0.62829
(±0.00112)
Rank: 89/113
0.51408
Rank: 92/113
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
[sid:00555] SuperPoint+GIFT+Graph Motion...
Size: 512 bytes. Matches: custom
1940.85 286.50
Rank: 2/52
0.356
Rank: 14/52
0.676
Rank: 43/52
0.42283
(±0.00000)
Rank: 14/52
290.06
Rank: 11/52
1887.70
Rank: 4/52
4.567
Rank: 6/52
0.420
Rank: 2/52
0.70395
(±0.00239)
Rank: 2/52
0.56339
Rank: 4/52
Anonymous (released after deadline) superpoint-2k scale-gift (128 float32: 512 bytes) Anonymous (released after deadline) N/A Anonymous (released after deadline) 20-05-10 is_baseline
[sid:00038] SOSNet, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 137.72
Rank: 24/52
0.323
Rank: 33/52
0.760
Rank: 32/52
0.38654
(±0.00056)
Rank: 26/52
145.47
Rank: 45/52
1271.41
Rank: 32/52
4.045
Rank: 33/52
0.526
Rank: 32/52
0.56072
(±0.00077)
Rank: 32/52
0.47363
Rank: 29/52
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
[sid:00007] L2-Net, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 89.22
Rank: 45/52
0.323
Rank: 39/52
0.779
Rank: 23/52
0.36265
(±0.00090)
Rank: 34/52
138.97
Rank: 47/52
1211.49
Rank: 37/52
3.948
Rank: 40/52
0.537
Rank: 35/52
0.52019
(±0.00120)
Rank: 38/52
0.44142
Rank: 38/52
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
[sid:00516] Key.Net (MS 4-1, Pyramid) + ...
Size: 512 bytes. Matches: built-in
2039.11 271.97
Rank: 3/52
0.447
Rank: 4/52
0.776
Rank: 25/52
0.40559
(±0.00112)
Rank: 17/52
336.81
Rank: 6/52
1559.74
Rank: 13/52
4.536
Rank: 7/52
0.461
Rank: 10/52
0.66406
(±0.00521)
Rank: 7/52
0.53483
Rank: 11/52
Anonymous (released after deadline) keynet-41-pyramidv2 x-net-lib (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-24 is_submission
[sid:00002] SURF (OpenCV), DEGENSAC
Size: 256 bytes. Matches: built-in
2010.88 73.55
Rank: 48/52
0.321
Rank: 45/52
0.634
Rank: 44/52
0.20761
(±0.00029)
Rank: 45/52
300.02
Rank: 10/52
1208.06
Rank: 39/52
3.549
Rank: 45/52
0.695
Rank: 45/52
0.37845
(±0.00399)
Rank: 45/52
0.29303
Rank: 45/52
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
[sid:00544] Key.Net + X-Net(Lib) w/ DEGE...
Size: 512 bytes. Matches: built-in
2039.11 198.77
Rank: 5/52
0.447
Rank: 4/52
0.814
Rank: 6/52
0.42332
(±0.00096)
Rank: 13/52
312.61
Rank: 8/52
1515.67
Rank: 14/52
4.567
Rank: 5/52
0.448
Rank: 4/52
0.66559
(±0.00233)
Rank: 6/52
0.54445
Rank: 5/52
Anonymous (released after deadline) keynet x-net-lib (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-28 is_submission
[sid:00523] SEKD
Size: 512 bytes. Matches: built-in
2043.44 119.88
Rank: 28/52
0.386
Rank: 10/52
0.799
Rank: 13/52
0.42811
(±0.00096)
Rank: 11/52
166.41
Rank: 35/52
1163.31
Rank: 41/52
4.427
Rank: 10/52
0.456
Rank: 7/52
0.64919
(±0.00171)
Rank: 8/52
0.53865
Rank: 8/52
Anonymous (released after deadline) sekd sekd (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-29 is_submission
[sid:00575] SIFT2k-2048-HardNet64-rain-a...
Size: 512 bytes. Matches: built-in
2047.77 165.12
Rank: 11/52
0.339
Rank: 16/52
0.814
Rank: 4/52
0.43543
(±0.00051)
Rank: 7/52
255.45
Rank: 15/52
1708.29
Rank: 8/52
4.327
Rank: 13/52
0.451
Rank: 6/52
0.64653
(±0.00221)
Rank: 9/52
0.54098
Rank: 6/52
Anonymous (released after deadline) sift2k sift2k-2048-hardnet64-train-all-sos-812000 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-20 is_baseline
[sid:00001] HardNet, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 112.03
Rank: 36/52
0.323
Rank: 33/52
0.776
Rank: 24/52
0.38578
(±0.00131)
Rank: 28/52
153.29
Rank: 40/52
1306.64
Rank: 28/52
4.026
Rank: 35/52
0.539
Rank: 38/52
0.55573
(±0.00175)
Rank: 34/52
0.47075
Rank: 31/52
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
[sid:00565] SIFT2k_2000_HardNet64-train-...
Size: 512 bytes. Matches: built-in
1999.83 160.96
Rank: 13/52
0.336
Rank: 17/52
0.813
Rank: 7/52
0.43188
(±0.00005)
Rank: 8/52
248.91
Rank: 17/52
1666.27
Rank: 10/52
4.323
Rank: 14/52
0.465
Rank: 13/52
0.64404
(±0.00271)
Rank: 11/52
0.53796
Rank: 9/52
Anonymous (released after deadline) sift2k sift2k-2000-hardnet64-train-all-sos-812000 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-13 is_baseline
[sid:00011] GeoDesc, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 95.61
Rank: 43/52
0.323
Rank: 39/52
0.759
Rank: 33/52
0.34215
(±0.00054)
Rank: 37/52
120.68
Rank: 50/52
1098.70
Rank: 45/52
4.002
Rank: 37/52
0.552
Rank: 40/52
0.50594
(±0.00073)
Rank: 40/52
0.42405
Rank: 40/52
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
[sid:00043] DELF-GLD (128D), PyRANSAC
Size: 512 bytes. Matches: built-in
2036.82 96.75
Rank: 42/52
0.109
Rank: 48/52
0.156
Rank: 48/52
0.07298
(±0.00056)
Rank: 50/52
430.68
Rank: 3/52
2162.72
Rank: 1/52
2.535
Rank: 48/52
0.887
Rank: 49/52
0.13247
(±0.00116)
Rank: 47/52
0.10272
Rank: 48/52
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
[sid:00578] Key.Net-s + DescNet w/ DEGEN...
Size: 512 bytes. Matches: built-in
2033.67 252.57
Rank: 4/52
0.449
Rank: 2/52
0.805
Rank: 10/52
0.45418
(±0.00056)
Rank: 3/52
331.55
Rank: 7/52
1621.68
Rank: 12/52
4.570
Rank: 4/52
0.447
Rank: 3/52
0.67414
(±0.00371)
Rank: 4/52
0.56416
Rank: 3/52
Anonymous (released after deadline) keynet-s descnet (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-20 is_baseline
[sid:00030] Upright Root-SIFT (OpenCV), ...
Size: 512 bytes. Matches: built-in
1892.72 114.22
Rank: 34/52
0.333
Rank: 19/52
0.781
Rank: 22/52
0.39547
(±0.00027)
Rank: 21/52
201.15
Rank: 22/52
1347.75
Rank: 24/52
4.091
Rank: 29/52
0.522
Rank: 29/52
0.56078
(±0.00264)
Rank: 31/52
0.47812
Rank: 26/52
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
[sid:00009] GeoDesc (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 119.25
Rank: 30/52
0.333
Rank: 21/52
0.784
Rank: 19/52
0.38660
(±0.00078)
Rank: 25/52
149.66
Rank: 43/52
1237.96
Rank: 34/52
4.219
Rank: 25/52
0.525
Rank: 31/52
0.56687
(±0.00261)
Rank: 29/52
0.47673
Rank: 27/52
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
[sid:00017] ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
2031.83 64.72
Rank: 51/52
0.366
Rank: 12/52
0.523
Rank: 47/52
0.08243
(±0.00058)
Rank: 48/52
120.93
Rank: 49/52
280.38
Rank: 52/52
2.191
Rank: 52/52
0.863
Rank: 47/52
0.02318
(±0.00229)
Rank: 52/52
0.05280
Rank: 52/52
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
[sid:00589] SuperPoint-128d + SuperGlue ...
Size: 512 bytes. Matches: custom
1973.62 325.80
Rank: 1/52
0.364
Rank: 13/52
0.772
Rank: 26/52
0.55214
(±0.00098)
Rank: 1/52
429.49
Rank: 5/52
2130.73
Rank: 3/52
4.570
Rank: 3/52
0.384
Rank: 1/52
0.75283
(±0.00086)
Rank: 1/52
0.65248
Rank: 1/52
Anonymous (released after deadline) superpoint-k2048-nms3-refine2-r1600forcecubic superpoint-down128 (128 float32: 512 bytes) Anonymous (released after deadline) https://arxiv.org/abs/1911.11763 Anonymous (released after deadline) 20-05-24 is_baseline
[sid:00036] SOSNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 159.24
Rank: 15/52
0.333
Rank: 21/52
0.794
Rank: 16/52
0.43077
(±0.00017)
Rank: 9/52
183.92
Rank: 26/52
1403.66
Rank: 19/52
4.314
Rank: 15/52
0.472
Rank: 16/52
0.62181
(±0.00161)
Rank: 19/52
0.52629
Rank: 17/52
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
[sid:00012] GeoDesc (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 148.30
Rank: 20/52
0.333
Rank: 21/52
0.765
Rank: 30/52
0.38124
(±0.00056)
Rank: 31/52
149.66
Rank: 43/52
1237.96
Rank: 34/52
4.219
Rank: 25/52
0.526
Rank: 33/52
0.56687
(±0.00261)
Rank: 29/52
0.47405
Rank: 28/52
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
[sid:00026] AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
1657.83 70.81
Rank: 50/52
0.355
Rank: 15/52
0.622
Rank: 45/52
0.17632
(±0.00087)
Rank: 46/52
224.40
Rank: 19/52
795.54
Rank: 51/52
3.311
Rank: 46/52
0.819
Rank: 46/52
0.23020
(±0.00466)
Rank: 46/52
0.20326
Rank: 46/52
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
[sid:00583] R2D2
Size: 512 bytes. Matches: built-in
2048.00 172.33
Rank: 10/52
0.398
Rank: 9/52
0.695
Rank: 42/52
0.33715
(±0.00149)
Rank: 39/52
306.15
Rank: 9/52
1235.83
Rank: 36/52
4.156
Rank: 28/52
0.496
Rank: 22/52
0.60223
(±0.00103)
Rank: 24/52
0.46969
Rank: 33/52
(contact) r2d2-5k-p-aug r2d2-5k-p-aug (128 float32: 512 bytes) N/A N/A 20-05-23 is_baseline
[sid:00513] Personal Inplementation SIFT...
Size: 512 bytes. Matches: built-in
1900.60 70.91
Rank: 49/52
0.085
Rank: 52/52
0.002
Rank: 52/52
0.07782
(±0.00022)
Rank: 49/52
173.06
Rank: 29/52
948.78
Rank: 50/52
3.816
Rank: 42/52
0.887
Rank: 50/52
0.09835
(±0.00068)
Rank: 49/52
0.08809
Rank: 49/52
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
[sid:00013] SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1936.30 78.65
Rank: 47/52
0.323
Rank: 43/52
0.733
Rank: 39/52
0.30766
(±0.00075)
Rank: 43/52
171.92
Rank: 32/52
1059.70
Rank: 49/52
3.687
Rank: 44/52
0.618
Rank: 43/52
0.42476
(±0.00103)
Rank: 43/52
0.36621
Rank: 43/52
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
[sid:00123] KeyNet-HardNet-2k
Size: 512 bytes. Matches: built-in
2048.00 137.55
Rank: 25/52
0.469
Rank: 1/52
0.818
Rank: 3/52
0.32725
(±0.00026)
Rank: 41/52
195.33
Rank: 23/52
1276.30
Rank: 31/52
4.493
Rank: 8/52
0.490
Rank: 21/52
0.61606
(±0.00141)
Rank: 21/52
0.47166
Rank: 30/52
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
[sid:00044] Upright SIFT (OpenCV), DEGEN...
Size: 512 bytes. Matches: built-in
1892.72 106.58
Rank: 39/52
0.333
Rank: 19/52
0.763
Rank: 31/52
0.37078
(±0.00056)
Rank: 33/52
205.17
Rank: 21/52
1300.49
Rank: 30/52
4.000
Rank: 39/52
0.563
Rank: 41/52
0.52549
(±0.00120)
Rank: 37/52
0.44814
Rank: 37/52
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
[sid:00503] Guided matching hardnet
Size: 512 bytes. Matches: custom
1892.71 152.91
Rank: 18/52
0.333
Rank: 30/52
0.795
Rank: 15/52
0.42159
(±0.00000)
Rank: 15/52
254.13
Rank: 16/52
1754.99
Rank: 7/52
4.283
Rank: 21/52
0.464
Rank: 12/52
0.64144
(±0.00207)
Rank: 13/52
0.53152
Rank: 13/52
Anonymous (released after deadline) sift upright-hardnet (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00022] L2-Net (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 113.44
Rank: 35/52
0.333
Rank: 21/52
0.801
Rank: 12/52
0.40729
(±0.00097)
Rank: 16/52
171.93
Rank: 30/52
1333.56
Rank: 25/52
4.226
Rank: 23/52
0.497
Rank: 24/52
0.58762
(±0.00228)
Rank: 25/52
0.49746
Rank: 21/52
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
[sid:00016] LogPolarDesc, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 151.94
Rank: 19/52
0.323
Rank: 33/52
0.752
Rank: 37/52
0.38840
(±0.00064)
Rank: 24/52
162.84
Rank: 37/52
1385.60
Rank: 22/52
4.046
Rank: 31/52
0.516
Rank: 27/52
0.57149
(±0.00344)
Rank: 27/52
0.47994
Rank: 25/52
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
[sid:00032] Root-SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1936.30 85.35
Rank: 46/52
0.323
Rank: 43/52
0.756
Rank: 36/52
0.33605
(±0.00092)
Rank: 40/52
167.84
Rank: 34/52
1141.88
Rank: 43/52
3.788
Rank: 43/52
0.576
Rank: 42/52
0.46826
(±0.00154)
Rank: 42/52
0.40216
Rank: 42/52
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
[sid:00040] DELF-GLD (128D), DEGENSAC
Size: 512 bytes. Matches: built-in
2036.82 103.04
Rank: 40/52
0.109
Rank: 48/52
0.151
Rank: 49/52
0.08367
(±0.00047)
Rank: 47/52
430.68
Rank: 3/52
2162.72
Rank: 1/52
2.535
Rank: 48/52
0.880
Rank: 48/52
0.13247
(±0.00116)
Rank: 47/52
0.10807
Rank: 47/52
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
[sid:00027] DELF-GLD (32D), PyRANSAC
Size: 128 bytes. Matches: built-in
2036.82 92.99
Rank: 44/52
0.109
Rank: 48/52
0.122
Rank: 50/52
0.06197
(±0.00032)
Rank: 52/52
453.19
Rank: 1/52
1805.48
Rank: 5/52
2.425
Rank: 50/52
0.916
Rank: 51/52
0.09163
(±0.00140)
Rank: 50/52
0.07680
Rank: 51/52
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
[sid:00037] GeoDesc, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 117.86
Rank: 32/52
0.323
Rank: 39/52
0.742
Rank: 38/52
0.34024
(±0.00050)
Rank: 38/52
120.68
Rank: 50/52
1098.70
Rank: 45/52
4.002
Rank: 37/52
0.551
Rank: 39/52
0.50594
(±0.00073)
Rank: 40/52
0.42309
Rank: 41/52
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
[sid:00023] LogPolarDesc, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 119.51
Rank: 29/52
0.323
Rank: 33/52
0.770
Rank: 27/52
0.39335
(±0.00060)
Rank: 22/52
162.84
Rank: 37/52
1385.60
Rank: 22/52
4.046
Rank: 31/52
0.519
Rank: 28/52
0.57149
(±0.00344)
Rank: 27/52
0.48242
Rank: 23/52
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
[sid:00564] SIFT2k_2000_HardNet64_train_...
Size: 512 bytes. Matches: built-in
1999.83 160.13
Rank: 14/52
0.336
Rank: 17/52
0.814
Rank: 5/52
0.42959
(±0.00060)
Rank: 10/52
244.13
Rank: 18/52
1646.76
Rank: 11/52
4.334
Rank: 12/52
0.459
Rank: 9/52
0.64407
(±0.00339)
Rank: 10/52
0.53683
Rank: 10/52
Anonymous (released after deadline) sift2k sift2k-2000-hardnet64-train-all-l2-138000 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-13 is_baseline
[sid:00025] DELF-GLD (32D), DEGENSAC
Size: 128 bytes. Matches: built-in
2036.82 98.57
Rank: 41/52
0.109
Rank: 48/52
0.119
Rank: 51/52
0.06573
(±0.00030)
Rank: 51/52
453.19
Rank: 1/52
1805.48
Rank: 5/52
2.425
Rank: 50/52
0.920
Rank: 52/52
0.09163
(±0.00140)
Rank: 50/52
0.07868
Rank: 50/52
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
[sid:00512] Hardnet with DEGENSAC for st...
Size: 512 bytes. Matches: built-in
1999.50 52.95
Rank: 52/52
0.172
Rank: 47/52
0.558
Rank: 46/52
0.20771
(±0.00049)
Rank: 44/52
181.48
Rank: 28/52
1692.63
Rank: 9/52
3.050
Rank: 47/52
0.695
Rank: 44/52
0.38463
(±0.00275)
Rank: 44/52
0.29617
Rank: 44/52
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
[sid:00527] SEKD
Size: 512 bytes. Matches: built-in
1786.69 108.69
Rank: 38/52
0.383
Rank: 11/52
0.759
Rank: 34/52
0.38886
(±0.00013)
Rank: 23/52
150.11
Rank: 42/52
1063.08
Rank: 48/52
4.575
Rank: 2/52
0.465
Rank: 14/52
0.63303
(±0.00244)
Rank: 14/52
0.51095
Rank: 20/52
Anonymous (released after deadline) sekd sekd (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-27 is_submission
[sid:00586] KeyNet+HardNet+NM-Net_v2
Size: 512 bytes. Matches: custom
2042.89 163.56
Rank: 12/52
0.435
Rank: 6/52
0.821
Rank: 1/52
0.40264
(±0.00000)
Rank: 18/52
165.01
Rank: 36/52
1134.69
Rank: 44/52
4.720
Rank: 1/52
0.448
Rank: 5/52
0.67488
(±0.00197)
Rank: 3/52
0.53876
Rank: 7/52
Anonymous (released after deadline) keynet hardnet (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-23 is_baseline
[sid:00585] NM-Net_v2
Size: 512 bytes. Matches: custom
1892.70 154.75
Rank: 17/52
0.333
Rank: 31/52
0.809
Rank: 8/52
0.46620
(±0.00000)
Rank: 2/52
155.91
Rank: 39/52
1392.47
Rank: 21/52
4.424
Rank: 11/52
0.459
Rank: 8/52
0.66837
(±0.00125)
Rank: 5/52
0.56729
Rank: 2/52
Anonymous (released after deadline) siftdef hardnet (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-23 is_baseline
[sid:00517] retrained Hardnet with MAGSA...
Size: 512 bytes. Matches: built-in
1892.71 136.55
Rank: 26/52
0.333
Rank: 21/52
0.808
Rank: 9/52
0.40042
(±0.00077)
Rank: 19/52
210.10
Rank: 20/52
1480.67
Rank: 15/52
4.226
Rank: 22/52
0.479
Rank: 20/52
0.63030
(±0.00242)
Rank: 15/52
0.51536
Rank: 18/52
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
[sid:00509] Upright-Sift + X-Net-lib w/ ...
Size: 512 bytes. Matches: built-in
1892.70 119.25
Rank: 31/52
0.333
Rank: 32/52
0.821
Rank: 2/52
0.42382
(±0.00089)
Rank: 12/52
168.67
Rank: 33/52
1319.63
Rank: 27/52
4.303
Rank: 17/52
0.468
Rank: 15/52
0.62953
(±0.00216)
Rank: 16/52
0.52668
Rank: 16/52
Anonymous (released after deadline) sift-def x-net-lib-upright-no-dups (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
CV-DoG-HardNet8-PT
Size: 512 bytes. Matches: built-in
2048.00 117.11
Rank: 33/52
0.321
Rank: 46/52
0.790
Rank: 17/52
0.38200
(±0.00081)
Rank: 30/52
117.84
Rank: 52/52
1067.13
Rank: 47/52
4.051
Rank: 30/52
0.531
Rank: 34/52
0.53035
(±0.00531)
Rank: 36/52
0.45617
Rank: 36/52
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_baseline
[sid:00003] HardNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 144.79
Rank: 21/52
0.333
Rank: 21/52
0.803
Rank: 11/52
0.44235
(±0.00127)
Rank: 4/52
193.09
Rank: 24/52
1435.93
Rank: 17/52
4.302
Rank: 18/52
0.474
Rank: 18/52
0.62275
(±0.00423)
Rank: 17/52
0.53255
Rank: 12/52
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
[sid:00035] HardNet (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 185.87
Rank: 6/52
0.333
Rank: 21/52
0.781
Rank: 21/52
0.43816
(±0.00058)
Rank: 5/52
193.09
Rank: 24/52
1435.93
Rank: 17/52
4.302
Rank: 18/52
0.473
Rank: 17/52
0.62275
(±0.00423)
Rank: 17/52
0.53045
Rank: 14/52
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
[sid:00004] SOSNet (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 180.53
Rank: 7/52
0.333
Rank: 21/52
0.782
Rank: 20/52
0.43803
(±0.00092)
Rank: 6/52
183.92
Rank: 26/52
1403.66
Rank: 19/52
4.314
Rank: 15/52
0.474
Rank: 19/52
0.62181
(±0.00161)
Rank: 19/52
0.52992
Rank: 15/52
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
[sid:00018] R2D2 (r2d2-wasf-n16), DEGENS...
Size: 512 bytes. Matches: built-in
2048.00 179.83
Rank: 8/52
0.422
Rank: 7/52
0.725
Rank: 40/52
0.35783
(±0.00061)
Rank: 35/52
262.02
Rank: 14/52
1188.11
Rank: 40/52
4.299
Rank: 20/52
0.499
Rank: 25/52
0.60639
(±0.00124)
Rank: 22/52
0.48211
Rank: 24/52
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
[sid:00014] L2-Net, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 110.16
Rank: 37/52
0.323
Rank: 39/52
0.766
Rank: 29/52
0.35649
(±0.00110)
Rank: 36/52
138.97
Rank: 47/52
1211.49
Rank: 37/52
3.948
Rank: 40/52
0.538
Rank: 37/52
0.52019
(±0.00120)
Rank: 38/52
0.43834
Rank: 39/52
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
[sid:00033] SOSNet, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 122.67
Rank: 27/52
0.323
Rank: 33/52
0.769
Rank: 28/52
0.37632
(±0.00056)
Rank: 32/52
145.47
Rank: 45/52
1271.41
Rank: 32/52
4.045
Rank: 33/52
0.524
Rank: 30/52
0.56072
(±0.00077)
Rank: 32/52
0.46852
Rank: 34/52
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
[sid:00039] HardNet, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 141.63
Rank: 23/52
0.323
Rank: 33/52
0.757
Rank: 35/52
0.38490
(±0.00120)
Rank: 29/52
153.29
Rank: 40/52
1306.64
Rank: 28/52
4.026
Rank: 35/52
0.537
Rank: 36/52
0.55573
(±0.00175)
Rank: 34/52
0.47032
Rank: 32/52
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
[sid:00582] R2D2
Size: 512 bytes. Matches: built-in
2048.00 157.87
Rank: 16/52
0.403
Rank: 8/52
0.709
Rank: 41/52
0.32501
(±0.00085)
Rank: 42/52
285.11
Rank: 12/52
1162.94
Rank: 42/52
4.161
Rank: 27/52
0.508
Rank: 26/52
0.60276
(±0.00062)
Rank: 23/52
0.46389
Rank: 35/52
(contact) r2d2-5k-p r2d2-5k-p (128 float32: 512 bytes) N/A N/A 20-05-23 is_baseline
[sid:00008] L2-Net (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 141.92
Rank: 22/52
0.333
Rank: 21/52
0.786
Rank: 18/52
0.39980
(±0.00020)
Rank: 20/52
171.93
Rank: 30/52
1333.56
Rank: 25/52
4.226
Rank: 23/52
0.496
Rank: 23/52
0.58762
(±0.00228)
Rank: 25/52
0.49371
Rank: 22/52
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
[sid:00531] Key.Net + X-Net w/ DEGENSAC
Size: 512 bytes. Matches: built-in
2040.08 177.80
Rank: 9/52
0.447
Rank: 3/52
0.797
Rank: 14/52
0.38608
(±0.00068)
Rank: 27/52
276.91
Rank: 13/52
1468.50
Rank: 16/52
4.447
Rank: 9/52
0.463
Rank: 11/52
0.64369
(±0.00081)
Rank: 12/52
0.51489
Rank: 19/52
Anonymous (released after deadline) keynet x-net (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-24 is_submission

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
[sid:00042] ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
7150.21 165.24
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
[sid:00010] AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
7857.11 251.77
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
[sid:00017] ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
2031.83 64.72
Rank: 4/4
0.366
Rank: 1/4
0.523
Rank: 2/4
0.08243
(±0.00058)
Rank: 2/4
120.93
Rank: 4/4
280.38
Rank: 4/4
2.191
Rank: 4/4
0.863
Rank: 2/4
0.02318
(±0.00229)
Rank: 4/4
0.05280
Rank: 4/4
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
[sid:00026] AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
1657.83 70.81
Rank: 3/4
0.355
Rank: 2/4
0.622
Rank: 1/4
0.17632
(±0.00087)
Rank: 1/4
224.40
Rank: 3/4
795.54
Rank: 3/4
3.311
Rank: 1/4
0.819
Rank: 1/4
0.23020
(±0.00466)
Rank: 1/4
0.20326
Rank: 1/4
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
[sid:00027] DELF-GLD (32D), PyRANSAC
Size: 128 bytes. Matches: built-in
2036.82 92.99
Rank: 2/4
0.109
Rank: 3/4
0.122
Rank: 3/4
0.06197
(±0.00032)
Rank: 4/4
453.19
Rank: 1/4
1805.48
Rank: 1/4
2.425
Rank: 2/4
0.916
Rank: 3/4
0.09163
(±0.00140)
Rank: 2/4
0.07680
Rank: 3/4
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
[sid:00025] DELF-GLD (32D), DEGENSAC
Size: 128 bytes. Matches: built-in
2036.82 98.57
Rank: 1/4
0.109
Rank: 3/4
0.119
Rank: 4/4
0.06573
(±0.00030)
Rank: 3/4
453.19
Rank: 1/4
1805.48
Rank: 1/4
2.425
Rank: 2/4
0.920
Rank: 4/4
0.09163
(±0.00140)
Rank: 2/4
0.07868
Rank: 2/4
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

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
[sid:00078] SOSNet-Upright w/ MAGSAC (no...
Size: 512 bytes. Matches: built-in
7829.63 693.71
Rank: 14/116
0.486
Rank: 36/116
0.855
Rank: 25/116
0.56626
(±0.00020)
Rank: 14/116
600.23
Rank: 25/116
4765.43
Rank: 17/116
4.584
Rank: 18/116
0.403
Rank: 17/116
0.73027
(±0.00284)
Rank: 15/116
0.64826
Rank: 16/116
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
[sid:00597] SIFT8k-HardNet64
Size: 512 bytes. Matches: built-in
6589.88 390.09
Rank: 70/116
0.467
Rank: 98/116
0.852
Rank: 30/116
0.52511
(±0.00056)
Rank: 42/116
420.42
Rank: 65/116
3595.93
Rank: 71/116
4.496
Rank: 37/116
0.417
Rank: 32/116
0.70738
(±0.00218)
Rank: 38/116
0.61625
Rank: 35/116
Anonymous (released after deadline) sift8k hardnet (128 float32: 512 bytes) Anonymous (released after deadline) https://arxiv.org/abs/1705.10872 Anonymous (released after deadline) 20-05-27 is_baseline
[sid:00042] ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
7150.21 165.24
Rank: 112/116
0.514
Rank: 15/116
0.653
Rank: 101/116
0.16159
(±0.00090)
Rank: 107/116
910.31
Rank: 11/116
1423.38
Rank: 114/116
2.722
Rank: 114/116
0.897
Rank: 116/116
0.08054
(±0.00242)
Rank: 116/116
0.12106
Rank: 116/116
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
[sid:00112] VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 209.62
Rank: 108/116
0.547
Rank: 11/116
0.823
Rank: 76/116
0.36954
(±0.00081)
Rank: 96/116
347.39
Rank: 97/116
3209.10
Rank: 91/116
4.126
Rank: 91/116
0.517
Rank: 100/116
0.58657
(±0.00120)
Rank: 99/116
0.47806
Rank: 98/116
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
[sid:00508] modified-r2d2
Size: 512 bytes. Matches: built-in
7791.82 163.86
Rank: 113/116
0.517
Rank: 14/116
0.785
Rank: 91/116
0.32815
(±0.00037)
Rank: 97/116
328.39
Rank: 103/116
2811.27
Rank: 102/116
4.250
Rank: 78/116
0.488
Rank: 92/116
0.62899
(±0.00193)
Rank: 88/116
0.47857
Rank: 97/116
Anonymous (released after deadline) modified-r2d2 modified-r2d2 (128 float32: 512 bytes) Anonymous (released after deadline) https://europe.naverlabs.com/wp-content/uploads/2019/09/R2D2-Repeatable-and-Reliable-Detector-and-Descriptor-2.pdf Anonymous (released after deadline) 20-04-25 is_submission
[sid:00536] HarrisZ + DeepOrientation + ...
Size: 512 bytes. Matches: custom
2410.18 1181.11
Rank: 2/116
0.368
Rank: 105/116
0.341
Rank: 114/116
0.00607
(±0.00000)
Rank: 115/116
1191.31
Rank: 3/116
2215.65
Rank: 106/116
4.014
Rank: 102/116
0.524
Rank: 102/116
0.58741
(±0.00250)
Rank: 98/116
0.29674
Rank: 108/116
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
[sid:00019] Upright Root-SIFT (OpenCV), ...
Size: 512 bytes. Matches: built-in
7829.24 363.31
Rank: 80/116
0.487
Rank: 21/116
0.844
Rank: 46/116
0.51009
(±0.00033)
Rank: 58/116
548.18
Rank: 30/116
4406.41
Rank: 36/116
4.385
Rank: 47/116
0.442
Rank: 58/116
0.68859
(±0.00254)
Rank: 59/116
0.59934
Rank: 57/116
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
[sid:00085] LogPolar-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 623.53
Rank: 17/116
0.486
Rank: 36/116
0.830
Rank: 65/116
0.51287
(±0.00038)
Rank: 55/116
483.93
Rank: 50/116
4405.57
Rank: 37/116
4.542
Rank: 29/116
0.423
Rank: 42/116
0.70680
(±0.00247)
Rank: 39/116
0.60983
Rank: 49/116
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
[sid:00571] ContextDesc Upright + OANetV...
Size: 512 bytes. Matches: custom
7830.09 843.99
Rank: 8/116
0.487
Rank: 23/116
0.775
Rank: 92/116
0.53810
(±0.00000)
Rank: 33/116
855.38
Rank: 12/116
6561.52
Rank: 2/116
4.522
Rank: 31/116
0.370
Rank: 4/116
0.76702
(±0.00081)
Rank: 2/116
0.65256
Rank: 13/116
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_baseline
[sid:00573] Guided-HardNet-qht
Size: 512 bytes. Matches: custom
7829.63 902.48
Rank: 6/116
0.486
Rank: 36/116
0.774
Rank: 93/116
0.55224
(±0.00000)
Rank: 21/116
538.28
Rank: 32/116
4693.85
Rank: 20/116
4.650
Rank: 5/116
0.381
Rank: 7/116
0.75930
(±0.00177)
Rank: 4/116
0.65577
Rank: 11/116
Anonymous (released after deadline) sift8k hardnet-qht (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-23 is_baseline
[sid:00530] SIFT-patch
Size: 1024 bytes. Matches: built-in
7861.62 85.40
Rank: 116/116
0.472
Rank: 74/116
0.577
Rank: 106/116
0.22977
(±0.00029)
Rank: 103/116
80.10
Rank: 116/116
973.09
Rank: 116/116
2.007
Rank: 115/116
0.726
Rank: 113/116
0.08184
(±0.00372)
Rank: 115/116
0.15581
Rank: 115/116
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 20-04-24 is_submission
[sid:00010] AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
7857.11 251.77
Rank: 103/116
0.553
Rank: 8/116
0.735
Rank: 99/116
0.30717
(±0.00122)
Rank: 100/116
479.55
Rank: 54/116
2778.68
Rank: 103/116
3.393
Rank: 111/116
0.737
Rank: 115/116
0.36048
(±0.00382)
Rank: 113/116
0.33383
Rank: 102/116
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
[sid:00068] GeoDesc-Upright w/ DEGENSAC ...
Size: 512 bytes. Matches: built-in
7829.63 418.12
Rank: 63/116
0.486
Rank: 36/116
0.869
Rank: 12/116
0.52674
(±0.00049)
Rank: 40/116
458.60
Rank: 55/116
4146.83
Rank: 48/116
4.412
Rank: 43/116
0.431
Rank: 50/116
0.70442
(±0.00189)
Rank: 42/116
0.61558
Rank: 36/116
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
[sid:00532] affnet-hardnet
Size: 512 bytes. Matches: built-in
7925.53 231.83
Rank: 105/116
0.452
Rank: 101/116
0.649
Rank: 103/116
0.31958
(±0.00055)
Rank: 98/116
243.28
Rank: 112/116
3155.72
Rank: 94/116
3.595
Rank: 107/116
0.554
Rank: 103/116
0.52578
(±0.00343)
Rank: 103/116
0.42268
Rank: 100/116
Anonymous (released after deadline) affnet hardnet (128 float32: 512 bytes) Anonymous (released after deadline) https://arxiv.org/abs/1711.06704 Anonymous (released after deadline) 20-04-24 is_submission
[sid:00071] L2-Net w/ DEGENSAC (no FLANN...
Size: 512 bytes. Matches: built-in
7861.11 374.25
Rank: 75/116
0.472
Rank: 78/116
0.852
Rank: 33/116
0.52953
(±0.00043)
Rank: 39/116
339.00
Rank: 101/116
3424.94
Rank: 80/116
4.206
Rank: 83/116
0.452
Rank: 72/116
0.66445
(±0.00213)
Rank: 73/116
0.59699
Rank: 60/116
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
[sid:00024] SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
7860.73 244.12
Rank: 104/116
0.472
Rank: 72/116
0.824
Rank: 74/116
0.45426
(±0.00097)
Rank: 87/116
418.86
Rank: 66/116
3515.63
Rank: 77/116
4.001
Rank: 103/116
0.502
Rank: 95/116
0.60193
(±0.00185)
Rank: 96/116
0.52810
Rank: 88/116
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
[sid:00084] LogPolar-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 468.77
Rank: 49/116
0.486
Rank: 36/116
0.852
Rank: 31/116
0.52316
(±0.00034)
Rank: 46/116
483.93
Rank: 50/116
4405.57
Rank: 37/116
4.542
Rank: 29/116
0.422
Rank: 39/116
0.70680
(±0.00247)
Rank: 39/116
0.61498
Rank: 37/116
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
[sid:00506] sift and hardnet64
Size: 512 bytes. Matches: built-in
7861.62 366.53
Rank: 79/116
0.472
Rank: 76/116
0.849
Rank: 36/116
0.52085
(±0.00009)
Rank: 47/116
585.33
Rank: 27/116
4957.86
Rank: 16/116
4.294
Rank: 70/116
0.419
Rank: 36/116
0.70472
(±0.00257)
Rank: 41/116
0.61279
Rank: 42/116
Anonymous (released after deadline) sift8k hardnet64 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00100] KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 456.82
Rank: 52/116
0.582
Rank: 1/116
0.838
Rank: 52/116
0.39971
(±0.00147)
Rank: 92/116
356.21
Rank: 88/116
3366.01
Rank: 83/116
4.319
Rank: 64/116
0.462
Rank: 80/116
0.64829
(±0.00174)
Rank: 81/116
0.52400
Rank: 89/116
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
[sid:00569] Upright-SIFT-HardNet
Size: 512 bytes. Matches: built-in
6667.86 384.82
Rank: 71/116
0.491
Rank: 17/116
0.844
Rank: 44/116
0.51560
(±0.00043)
Rank: 52/116
397.09
Rank: 75/116
3417.56
Rank: 82/116
4.479
Rank: 39/116
0.448
Rank: 71/116
0.68803
(±0.00532)
Rank: 60/116
0.60181
Rank: 55/116
Anonymous (released after deadline) sift8k hardnet (128 float32: 512 bytes) Anonymous (released after deadline) https://arxiv.org/abs/1705.10872 Anonymous (released after deadline) 20-05-20 is_baseline
[sid:00066] SOSNet-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 549.86
Rank: 32/116
0.486
Rank: 36/116
0.841
Rank: 49/116
0.53531
(±0.00061)
Rank: 35/116
521.37
Rank: 40/116
4473.48
Rank: 31/116
4.597
Rank: 15/116
0.414
Rank: 28/116
0.72179
(±0.00330)
Rank: 24/116
0.62855
Rank: 30/116
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
[sid:00076] SOSNet w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 575.53
Rank: 30/116
0.472
Rank: 78/116
0.846
Rank: 39/116
0.55170
(±0.00042)
Rank: 22/116
508.66
Rank: 45/116
4502.19
Rank: 28/116
4.377
Rank: 55/116
0.405
Rank: 23/116
0.71820
(±0.00298)
Rank: 29/116
0.63495
Rank: 24/116
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
[sid:00065] SOSNet-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 415.92
Rank: 64/116
0.486
Rank: 36/116
0.864
Rank: 21/116
0.54098
(±0.00086)
Rank: 29/116
521.37
Rank: 40/116
4473.48
Rank: 31/116
4.597
Rank: 15/116
0.417
Rank: 30/116
0.72179
(±0.00330)
Rank: 24/116
0.63139
Rank: 26/116
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
[sid:00522] SIFT and HardNet64 train sca...
Size: 512 bytes. Matches: built-in
7830.09 523.62
Rank: 39/116
0.486
Rank: 27/116
0.852
Rank: 34/116
0.55385
(±0.00076)
Rank: 20/116
822.07
Rank: 14/116
5660.19
Rank: 10/116
4.555
Rank: 24/116
0.404
Rank: 21/116
0.72744
(±0.00455)
Rank: 18/116
0.64065
Rank: 21/116
Anonymous (released after deadline) sift8k hardnet64-train-raw64 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-26 is_submission
[sid:00554] ContextDesc-Upright w/ DEGEN...
Size: 512 bytes. Matches: built-in
7830.09 480.10
Rank: 47/116
0.487
Rank: 23/116
0.877
Rank: 1/116
0.55697
(±0.00079)
Rank: 18/116
543.45
Rank: 31/116
4543.76
Rank: 27/116
4.498
Rank: 36/116
0.405
Rank: 22/116
0.72902
(±0.00255)
Rank: 17/116
0.64299
Rank: 20/116
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_baseline
[sid:00504] ASLFeat (MS) w/ DEGENSAC (no...
Size: 512 bytes. Matches: built-in
6948.83 396.95
Rank: 67/116
0.550
Rank: 9/116
0.797
Rank: 88/116
0.46102
(±0.00067)
Rank: 84/116
510.80
Rank: 42/116
3817.14
Rank: 64/116
4.577
Rank: 20/116
0.444
Rank: 63/116
0.68246
(±0.00173)
Rank: 63/116
0.57174
Rank: 76/116
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
[sid:00548] HarrisZ + DeepOrientation + ...
Size: 512 bytes. Matches: custom
2410.18 348.55
Rank: 84/116
0.368
Rank: 105/116
0.607
Rank: 105/116
0.28795
(±0.00000)
Rank: 101/116
351.53
Rank: 91/116
1699.08
Rank: 111/116
4.124
Rank: 94/116
0.507
Rank: 96/116
0.61370
(±0.00246)
Rank: 92/116
0.45082
Rank: 99/116
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
[sid:00519] SOSNet-Upright-AdaLAM
Size: 512 bytes. Matches: custom
7830.09 818.02
Rank: 10/116
0.486
Rank: 27/116
0.818
Rank: 80/116
0.56056
(±0.00000)
Rank: 16/116
827.85
Rank: 13/116
6157.62
Rank: 7/116
4.700
Rank: 3/116
0.381
Rank: 8/116
0.75917
(±0.00446)
Rank: 5/116
0.65987
Rank: 10/116
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. N/A N/A 20-04-26 is_submission
[sid:00101] KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 610.11
Rank: 23/116
0.582
Rank: 1/116
0.826
Rank: 71/116
0.49856
(±0.00038)
Rank: 64/116
356.21
Rank: 88/116
3366.01
Rank: 83/116
4.319
Rank: 64/116
0.464
Rank: 83/116
0.64829
(±0.00174)
Rank: 81/116
0.57342
Rank: 74/116
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
[sid:00501] Guided matching with Upright...
Size: 512 bytes. Matches: custom
7829.27 620.79
Rank: 19/116
0.486
Rank: 25/116
0.786
Rank: 90/116
0.49072
(±0.00000)
Rank: 68/116
1363.98
Rank: 2/116
7835.75
Rank: 1/116
4.219
Rank: 82/116
0.454
Rank: 74/116
0.68064
(±0.00390)
Rank: 66/116
0.58568
Rank: 66/116
Anonymous (released after deadline) sift upright-root-sift (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00060] L2-Net w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 396.66
Rank: 68/116
0.472
Rank: 78/116
0.813
Rank: 84/116
0.49054
(±0.00043)
Rank: 69/116
314.54
Rank: 107/116
3314.19
Rank: 89/116
4.204
Rank: 85/116
0.465
Rank: 84/116
0.65665
(±0.00287)
Rank: 75/116
0.57360
Rank: 73/116
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
[sid:00545]structured feature descriptio...
Size: 512 bytes. Matches: built-in
7830.21 420.34
Rank: 61/116
0.486
Rank: 34/116
0.865
Rank: 18/116
0.54442
(±0.00084)
Rank: 27/116
524.05
Rank: 38/116
4467.23
Rank: 33/116
4.551
Rank: 25/116
0.412
Rank: 25/116
0.72729
(±0.00300)
Rank: 19/116
0.63586
Rank: 23/116
Anonymous (released after deadline) sift sfd (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-29 is_submission
[sid:00120] CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 270.99
Rank: 99/116
0.472
Rank: 69/116
0.873
Rank: 9/116
0.46073
(±0.00062)
Rank: 85/116
356.56
Rank: 85/116
3550.63
Rank: 72/116
4.275
Rank: 72/116
0.439
Rank: 56/116
0.68875
(±0.00138)
Rank: 56/116
0.57474
Rank: 72/116
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
[sid:00568] Guided-SOSNet-lib-p
Size: 512 bytes. Matches: custom
7829.63 517.65
Rank: 41/116
0.486
Rank: 36/116
0.874
Rank: 8/116
0.57982
(±0.00000)
Rank: 10/116
524.66
Rank: 37/116
4618.86
Rank: 23/116
4.632
Rank: 8/116
0.369
Rank: 2/116
0.75888
(±0.00437)
Rank: 6/116
0.66935
Rank: 4/116
Anonymous (released after deadline) sift8k sosnet-lib-p (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-15 is_baseline
[sid:00051] HardNet w/ DEGENSAC (no FLAN...
Size: 512 bytes. Matches: built-in
7861.11 441.30
Rank: 58/116
0.472
Rank: 78/116
0.866
Rank: 17/116
0.55430
(±0.00031)
Rank: 19/116
426.84
Rank: 63/116
4001.40
Rank: 56/116
4.339
Rank: 62/116
0.419
Rank: 34/116
0.70962
(±0.00191)
Rank: 36/116
0.63196
Rank: 25/116
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
[sid:00518] UprightRootSIFT-AdaLAM
Size: 512 bytes. Matches: custom
6449.42 443.61
Rank: 57/116
0.436
Rank: 102/116
0.825
Rank: 73/116
0.49625
(±0.00000)
Rank: 65/116
449.49
Rank: 58/116
4353.56
Rank: 39/116
4.453
Rank: 42/116
0.397
Rank: 15/116
0.72490
(±0.00190)
Rank: 20/116
0.61057
Rank: 46/116
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. N/A N/A 20-04-24 is_submission
[sid:00058] HardNet-Upright w/ MAGSAC (n...
Size: 512 bytes. Matches: built-in
7829.63 722.10
Rank: 13/116
0.486
Rank: 36/116
0.853
Rank: 27/116
0.56374
(±0.00069)
Rank: 15/116
509.07
Rank: 43/116
4250.40
Rank: 40/116
4.548
Rank: 26/116
0.412
Rank: 26/116
0.72309
(±0.00141)
Rank: 22/116
0.64341
Rank: 19/116
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
[sid:00034] D2-Net (single-scale), DEGEN...
Size: 2048 bytes. Matches: built-in
5665.31 256.26
Rank: 101/116
0.288
Rank: 111/116
0.360
Rank: 113/116
0.18499
(±0.00022)
Rank: 106/116
740.01
Rank: 19/116
4584.07
Rank: 26/116
3.506
Rank: 108/116
0.696
Rank: 112/116
0.41726
(±0.00175)
Rank: 111/116
0.30112
Rank: 107/116
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
[sid:00122] CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 539.84
Rank: 34/116
0.472
Rank: 69/116
0.838
Rank: 51/116
0.53291
(±0.00047)
Rank: 37/116
356.56
Rank: 85/116
3550.63
Rank: 72/116
4.275
Rank: 72/116
0.443
Rank: 62/116
0.68875
(±0.00138)
Rank: 56/116
0.61083
Rank: 45/116
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
[sid:00061] L2-Net-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 366.70
Rank: 78/116
0.486
Rank: 36/116
0.845
Rank: 40/116
0.52012
(±0.00106)
Rank: 49/116
369.60
Rank: 83/116
3538.91
Rank: 75/116
4.407
Rank: 45/116
0.447
Rank: 68/116
0.68108
(±0.00177)
Rank: 64/116
0.60060
Rank: 56/116
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
[sid:00121] CV-DoG-HardNetAmos-8k
Size: 512 bytes. Matches: built-in
7860.98 406.63
Rank: 66/116
0.472
Rank: 69/116
0.863
Rank: 22/116
0.53852
(±0.00039)
Rank: 32/116
356.56
Rank: 85/116
3550.63
Rank: 72/116
4.275
Rank: 72/116
0.439
Rank: 55/116
0.68875
(±0.00138)
Rank: 56/116
0.61364
Rank: 39/116
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
[sid:00069] GeoDesc w/ DEGENSAC (no FLAN...
Size: 512 bytes. Matches: built-in
7861.11 356.12
Rank: 83/116
0.472
Rank: 78/116
0.856
Rank: 24/116
0.51112
(±0.00070)
Rank: 57/116
395.12
Rank: 78/116
3838.97
Rank: 62/116
4.264
Rank: 75/116
0.443
Rank: 60/116
0.68032
(±0.00099)
Rank: 67/116
0.59572
Rank: 62/116
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
[sid:00591] Guided-Hardnet-Epoch2
Size: 512 bytes. Matches: custom
7829.63 774.81
Rank: 11/116
0.486
Rank: 36/116
0.823
Rank: 77/116
0.61081
(±0.00000)
Rank: 1/116
785.57
Rank: 17/116
6330.70
Rank: 4/116
4.680
Rank: 4/116
0.358
Rank: 1/116
0.78288
(±0.00094)
Rank: 1/116
0.69684
Rank: 1/116
Anonymous (released after deadline) sift8k hardnet-epoch2 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-24 is_baseline
[sid:00086] LogPolar-Upright w/ MAGSAC (...
Size: 512 bytes. Matches: built-in
7829.63 746.24
Rank: 12/116
0.486
Rank: 36/116
0.844
Rank: 45/116
0.54060
(±0.00047)
Rank: 30/116
505.37
Rank: 48/116
4414.11
Rank: 34/116
4.518
Rank: 33/116
0.421
Rank: 38/116
0.71092
(±0.00251)
Rank: 33/116
0.62576
Rank: 33/116
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
[sid:00106] VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 333.93
Rank: 89/116
0.490
Rank: 18/116
0.809
Rank: 85/116
0.46326
(±0.00089)
Rank: 82/116
324.62
Rank: 104/116
3030.67
Rank: 98/116
4.173
Rank: 87/116
0.462
Rank: 81/116
0.62829
(±0.00112)
Rank: 89/116
0.54578
Rank: 84/116
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
[sid:00081] LogPolar w/ MAGSAC (no FLANN...
Size: 512 bytes. Matches: built-in
7861.11 603.70
Rank: 24/116
0.472
Rank: 78/116
0.832
Rank: 61/116
0.52385
(±0.00113)
Rank: 44/116
415.77
Rank: 68/116
4054.60
Rank: 51/116
4.316
Rank: 67/116
0.432
Rank: 51/116
0.69284
(±0.00429)
Rank: 51/116
0.60835
Rank: 50/116
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
[sid:00562] HardNet64-train-all-raw64-ba...
Size: 512 bytes. Matches: built-in
7831.92 377.01
Rank: 74/116
0.486
Rank: 64/116
0.807
Rank: 86/116
0.45770
(±0.00060)
Rank: 86/116
575.43
Rank: 28/116
4756.46
Rank: 19/116
4.563
Rank: 23/116
0.435
Rank: 54/116
0.69780
(±0.00191)
Rank: 47/116
0.57775
Rank: 70/116
Anonymous (released after deadline) sift8k hardnet64-train-all-raw64-balance-l2-224000-raw64 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-13 is_baseline
[sid:00557] HarrisZ + DeepOrientation + ...
Size: 512 bytes. Matches: custom
2410.18 622.70
Rank: 18/116
0.368
Rank: 105/116
0.513
Rank: 108/116
0.05482
(±0.00000)
Rank: 110/116
626.57
Rank: 22/116
1793.00
Rank: 110/116
4.314
Rank: 69/116
0.517
Rank: 99/116
0.56933
(±0.00268)
Rank: 102/116
0.31207
Rank: 105/116
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_baseline
[sid:00520] SIFT and HardNet64 train sca...
Size: 512 bytes. Matches: built-in
7830.09 408.60
Rank: 65/116
0.486
Rank: 27/116
0.837
Rank: 56/116
0.50589
(±0.00059)
Rank: 59/116
619.25
Rank: 23/116
4959.78
Rank: 15/116
4.384
Rank: 48/116
0.456
Rank: 76/116
0.69172
(±0.00048)
Rank: 53/116
0.59880
Rank: 58/116
Anonymous (released after deadline) sift8k hardnet64-train-scale5 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00526] ASLFeat-MS
Size: 512 bytes. Matches: built-in
7384.19 464.39
Rank: 50/116
0.578
Rank: 4/116
0.752
Rank: 98/116
0.42208
(±0.00054)
Rank: 90/116
532.83
Rank: 35/116
4002.80
Rank: 55/116
4.521
Rank: 32/116
0.447
Rank: 65/116
0.66568
(±0.00347)
Rank: 72/116
0.54388
Rank: 85/116
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. Under preparation N/A 20-04-27 is_submission
[sid:00546] HarrisZ + DeepOrientation + ...
Size: 512 bytes. Matches: custom
2410.18 1036.77
Rank: 5/116
0.368
Rank: 105/116
0.366
Rank: 112/116
0.07780
(±0.00000)
Rank: 109/116
1047.01
Rank: 6/116
2197.09
Rank: 107/116
4.023
Rank: 101/116
0.509
Rank: 97/116
0.59222
(±0.00276)
Rank: 97/116
0.33501
Rank: 101/116
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
[sid:00524] SIFT-patch
Size: 512 bytes. Matches: built-in
7861.62 85.68
Rank: 115/116
0.472
Rank: 74/116
0.574
Rank: 107/116
0.22805
(±0.00076)
Rank: 104/116
81.93
Rank: 115/116
991.42
Rank: 115/116
1.934
Rank: 116/116
0.687
Rank: 111/116
0.08662
(±0.00384)
Rank: 114/116
0.15733
Rank: 114/116
feyman_priv (contact) sift8k l2net-arcface (128 float32: 512 bytes) sift8k with l2net trained on google-landmark-dataset-v1(in 1000 class) N/A 20-04-26 is_submission
[sid:00082] LogPolar w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 507.35
Rank: 42/116
0.472
Rank: 78/116
0.817
Rank: 81/116
0.49129
(±0.00011)
Rank: 67/116
399.70
Rank: 73/116
4036.37
Rank: 53/116
4.340
Rank: 60/116
0.429
Rank: 49/116
0.69100
(±0.00154)
Rank: 54/116
0.59115
Rank: 65/116
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
[sid:00116] VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 214.47
Rank: 107/116
0.577
Rank: 5/116
0.841
Rank: 50/116
0.39330
(±0.00052)
Rank: 94/116
350.69
Rank: 92/116
3327.71
Rank: 86/116
4.076
Rank: 98/116
0.490
Rank: 94/116
0.60691
(±0.00338)
Rank: 93/116
0.50011
Rank: 96/116
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
[sid:00063] SOSNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 347.73
Rank: 85/116
0.472
Rank: 78/116
0.854
Rank: 26/116
0.51891
(±0.00067)
Rank: 50/116
440.21
Rank: 60/116
4178.82
Rank: 43/116
4.378
Rank: 53/116
0.419
Rank: 35/116
0.70078
(±0.00375)
Rank: 45/116
0.60984
Rank: 48/116
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
[sid:00109] VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 255.82
Rank: 102/116
0.482
Rank: 66/116
0.838
Rank: 53/116
0.46795
(±0.00075)
Rank: 78/116
311.54
Rank: 109/116
3061.54
Rank: 95/116
4.105
Rank: 95/116
0.475
Rank: 87/116
0.62964
(±0.00428)
Rank: 85/116
0.54880
Rank: 81/116
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
[sid:00118] VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 358.98
Rank: 81/116
0.577
Rank: 5/116
0.824
Rank: 75/116
0.46263
(±0.00079)
Rank: 83/116
350.69
Rank: 92/116
3327.71
Rank: 86/116
4.076
Rank: 98/116
0.489
Rank: 93/116
0.60691
(±0.00338)
Rank: 93/116
0.53477
Rank: 87/116
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
[sid:00576] Upright-Sift + X-Net-lib w/ ...
Size: 512 bytes. Matches: built-in
7830.16 475.12
Rank: 48/116
0.486
Rank: 35/116
0.844
Rank: 43/116
0.53156
(±0.00027)
Rank: 38/116
720.29
Rank: 20/116
5398.84
Rank: 12/116
4.547
Rank: 28/116
0.408
Rank: 24/116
0.71966
(±0.00160)
Rank: 28/116
0.62561
Rank: 34/116
Anonymous (released after deadline) sift x-net-lib (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-20 is_baseline
[sid:00075] SOSNet w/ DEGENSAC (no FLANN...
Size: 512 bytes. Matches: built-in
7861.11 433.37
Rank: 60/116
0.472
Rank: 78/116
0.868
Rank: 14/116
0.55867
(±0.00060)
Rank: 17/116
508.66
Rank: 45/116
4502.19
Rank: 28/116
4.377
Rank: 55/116
0.404
Rank: 20/116
0.71820
(±0.00298)
Rank: 29/116
0.63844
Rank: 22/116
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
[sid:00105] VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 267.54
Rank: 100/116
0.490
Rank: 18/116
0.826
Rank: 72/116
0.46555
(±0.00055)
Rank: 81/116
324.62
Rank: 104/116
3030.67
Rank: 98/116
4.173
Rank: 87/116
0.461
Rank: 79/116
0.62829
(±0.00112)
Rank: 89/116
0.54692
Rank: 83/116
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
[sid:00062] L2-Net-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 483.13
Rank: 46/116
0.486
Rank: 36/116
0.822
Rank: 78/116
0.51450
(±0.00058)
Rank: 53/116
369.60
Rank: 83/116
3538.91
Rank: 75/116
4.407
Rank: 45/116
0.448
Rank: 70/116
0.68108
(±0.00177)
Rank: 64/116
0.59779
Rank: 59/116
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
[sid:00048] GeoDesc w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 371.02
Rank: 77/116
0.472
Rank: 78/116
0.816
Rank: 82/116
0.46887
(±0.00025)
Rank: 77/116
350.43
Rank: 95/116
3601.67
Rank: 69/116
4.234
Rank: 79/116
0.447
Rank: 67/116
0.65553
(±0.00444)
Rank: 78/116
0.56220
Rank: 79/116
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
[sid:00056] HardNet-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 448.53
Rank: 55/116
0.486
Rank: 36/116
0.864
Rank: 20/116
0.54393
(±0.00027)
Rank: 28/116
480.13
Rank: 52/116
4167.34
Rank: 45/116
4.566
Rank: 21/116
0.417
Rank: 31/116
0.71542
(±0.00274)
Rank: 31/116
0.62967
Rank: 29/116
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
[sid:00049] GeoDesc-Upright w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7829.63 324.65
Rank: 92/116
0.486
Rank: 36/116
0.850
Rank: 35/116
0.48769
(±0.00064)
Rank: 71/116
394.67
Rank: 80/116
3863.42
Rank: 60/116
4.380
Rank: 51/116
0.440
Rank: 57/116
0.67682
(±0.00127)
Rank: 69/116
0.58225
Rank: 67/116
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
[sid:00057] HardNet-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 595.31
Rank: 26/116
0.486
Rank: 36/116
0.841
Rank: 48/116
0.53730
(±0.00082)
Rank: 34/116
480.13
Rank: 52/116
4167.34
Rank: 45/116
4.566
Rank: 21/116
0.416
Rank: 29/116
0.71542
(±0.00274)
Rank: 31/116
0.62636
Rank: 32/116
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
[sid:00047] GeoDesc w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 287.87
Rank: 96/116
0.472
Rank: 78/116
0.838
Rank: 54/116
0.47259
(±0.00033)
Rank: 76/116
350.43
Rank: 95/116
3601.67
Rank: 69/116
4.234
Rank: 79/116
0.447
Rank: 66/116
0.65553
(±0.00444)
Rank: 78/116
0.56406
Rank: 78/116
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
[sid:00558] SIFT + DeepOrientation + SOS...
Size: 512 bytes. Matches: custom
2826.82 339.17
Rank: 87/116
0.284
Rank: 112/116
0.427
Rank: 110/116
0.01806
(±0.00000)
Rank: 114/116
341.75
Rank: 100/116
2003.56
Rank: 108/116
3.735
Rank: 105/116
0.593
Rank: 104/116
0.51048
(±0.00436)
Rank: 104/116
0.26427
Rank: 110/116
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_baseline
[sid:00079] LogPolar w/ DEGENSAC (no FLA...
Size: 512 bytes. Matches: built-in
7861.11 450.82
Rank: 54/116
0.472
Rank: 78/116
0.852
Rank: 29/116
0.53396
(±0.00086)
Rank: 36/116
415.77
Rank: 68/116
4054.60
Rank: 51/116
4.316
Rank: 67/116
0.432
Rank: 53/116
0.69284
(±0.00429)
Rank: 51/116
0.61340
Rank: 40/116
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
[sid:00077] SOSNet-Upright w/ DEGENSAC (...
Size: 512 bytes. Matches: built-in
7829.63 518.08
Rank: 40/116
0.486
Rank: 36/116
0.877
Rank: 2/116
0.57385
(±0.00041)
Rank: 11/116
600.23
Rank: 25/116
4765.43
Rank: 17/116
4.584
Rank: 18/116
0.403
Rank: 18/116
0.73027
(±0.00284)
Rank: 15/116
0.65206
Rank: 14/116
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
[sid:00502] Guided matching with Upright...
Size: 512 bytes. Matches: custom
7829.27 501.50
Rank: 43/116
0.486
Rank: 25/116
0.828
Rank: 69/116
0.50310
(±0.01761)
Rank: 61/116
809.76
Rank: 15/116
5966.91
Rank: 8/116
4.462
Rank: 40/116
0.425
Rank: 45/116
0.71009
(±0.00414)
Rank: 35/116
0.60659
Rank: 53/116
Anonymous (released after deadline) sift upright-root-sift (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00015] Root-SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
7860.73 281.05
Rank: 98/116
0.472
Rank: 72/116
0.845
Rank: 41/116
0.48887
(±0.00011)
Rank: 70/116
437.74
Rank: 62/116
3814.80
Rank: 65/116
4.151
Rank: 90/116
0.458
Rank: 77/116
0.65067
(±0.00155)
Rank: 80/116
0.56977
Rank: 77/116
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
[sid:00542] sift and hardnet64 train sca...
Size: 512 bytes. Matches: built-in
7830.09 618.78
Rank: 20/116
0.486
Rank: 27/116
0.872
Rank: 10/116
0.58801
(±0.00037)
Rank: 3/116
950.74
Rank: 9/116
6286.36
Rank: 6/116
4.626
Rank: 9/116
0.390
Rank: 14/116
0.74662
(±0.00087)
Rank: 11/116
0.66732
Rank: 5/116
Anonymous (released after deadline) sift8k hardnet64-train-all-l2 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-01 is_submission
[sid:00535] hardnetlib32P-Upright w/ DEG...
Size: 512 bytes. Matches: built-in
7829.63 537.62
Rank: 35/116
0.486
Rank: 36/116
0.876
Rank: 4/116
0.57327
(±0.00078)
Rank: 12/116
506.73
Rank: 47/116
4230.98
Rank: 42/116
4.517
Rank: 35/116
0.418
Rank: 33/116
0.72171
(±0.00156)
Rank: 26/116
0.64749
Rank: 18/116
Anonymous (released after deadline) sift8k-no-dups-mps hardnetlib32P (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-27 is_submission
[sid:00114] VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 358.21
Rank: 82/116
0.547
Rank: 11/116
0.800
Rank: 87/116
0.43347
(±0.00084)
Rank: 89/116
347.39
Rank: 97/116
3209.10
Rank: 91/116
4.126
Rank: 91/116
0.517
Rank: 98/116
0.58657
(±0.00120)
Rank: 99/116
0.51002
Rank: 95/116
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
[sid:00563] HardNet64-train-all-SOS-8120...
Size: 512 bytes. Matches: built-in
7830.09 624.02
Rank: 16/116
0.486
Rank: 27/116
0.869
Rank: 13/116
0.58664
(±0.00095)
Rank: 4/116
972.47
Rank: 7/116
6369.03
Rank: 3/116
4.612
Rank: 11/116
0.388
Rank: 12/116
0.74668
(±0.00251)
Rank: 10/116
0.66666
Rank: 6/116
Anonymous (released after deadline) sift8k hardnet64-train-all-sos-812000 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-13 is_baseline
[sid:00055] HardNet-Upright w/ DEGENSAC ...
Size: 512 bytes. Matches: built-in
7829.63 537.62
Rank: 36/116
0.486
Rank: 36/116
0.876
Rank: 3/116
0.57279
(±0.00120)
Rank: 13/116
509.07
Rank: 43/116
4250.40
Rank: 40/116
4.548
Rank: 26/116
0.413
Rank: 27/116
0.72309
(±0.00141)
Rank: 22/116
0.64794
Rank: 17/116
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
[sid:00108] VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 175.46
Rank: 110/116
0.482
Rank: 66/116
0.848
Rank: 37/116
0.39839
(±0.00122)
Rank: 93/116
311.54
Rank: 109/116
3061.54
Rank: 95/116
4.105
Rank: 95/116
0.475
Rank: 89/116
0.62964
(±0.00428)
Rank: 85/116
0.51401
Rank: 93/116
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
CV-DoG-HardNet8-PT
Size: 512 bytes. Matches: built-in
7829.26 589.94
Rank: 28/116
0.484
Rank: 65/116
0.872
Rank: 11/116
0.58375
(±0.00046)
Rank: 6/116
573.22
Rank: 29/116
4489.23
Rank: 30/116
4.619
Rank: 10/116
0.403
Rank: 19/116
0.72432
(±0.00357)
Rank: 21/116
0.65403
Rank: 12/116
Milan Pultar, Dmytro Mishkin, Jiri Matas (contact) sift8k h8e512pt (128 float32: 512 bytes) [sid:00594] HardNet8 with PCA compression N/A N/A 20-05-27 is_baseline
[sid:00533] upright-sift8k-hardnet
Size: 512 bytes. Matches: built-in
6589.88 379.73
Rank: 73/116
0.467
Rank: 98/116
0.852
Rank: 32/116
0.52463
(±0.00070)
Rank: 43/116
404.02
Rank: 70/116
3501.62
Rank: 78/116
4.483
Rank: 38/116
0.428
Rank: 47/116
0.70123
(±0.00081)
Rank: 44/116
0.61293
Rank: 41/116
Anonymous (released after deadline) sift8k hardnet (128 float32: 512 bytes) Anonymous (released after deadline) https://arxiv.org/abs/1705.10872 Anonymous (released after deadline) 20-04-24 is_submission
[sid:00053] HardNet w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 587.65
Rank: 29/116
0.472
Rank: 78/116
0.842
Rank: 47/116
0.55022
(±0.00001)
Rank: 24/116
426.84
Rank: 63/116
4001.40
Rank: 56/116
4.339
Rank: 62/116
0.419
Rank: 37/116
0.70962
(±0.00191)
Rank: 36/116
0.62992
Rank: 28/116
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
[sid:00005] SURF (OpenCV), DEGENSAC
Size: 256 bytes. Matches: built-in
7728.57 128.86
Rank: 114/116
0.432
Rank: 104/116
0.684
Rank: 100/116
0.24948
(±0.00065)
Rank: 102/116
928.29
Rank: 10/116
3455.28
Rank: 79/116
3.428
Rank: 109/116
0.733
Rank: 114/116
0.40490
(±0.00448)
Rank: 112/116
0.32719
Rank: 103/116
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
[sid:00510] sift and hardnet64
Size: 512 bytes. Matches: built-in
7861.62 333.97
Rank: 88/116
0.472
Rank: 76/116
0.845
Rank: 42/116
0.51149
(±0.00018)
Rank: 56/116
521.58
Rank: 39/116
4587.88
Rank: 24/116
4.222
Rank: 81/116
0.424
Rank: 44/116
0.69458
(±0.00340)
Rank: 50/116
0.60304
Rank: 54/116
Anonymous (released after deadline) sift8k hardnet64-v2 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00559] SIFT + DeepOrientation + SOS...
Size: 512 bytes. Matches: custom
2826.82 166.47
Rank: 111/116
0.284
Rank: 112/116
0.650
Rank: 102/116
0.05206
(±0.00000)
Rank: 111/116
167.59
Rank: 114/116
1549.91
Rank: 112/116
3.989
Rank: 104/116
0.622
Rank: 106/116
0.47700
(±0.00167)
Rank: 106/116
0.26453
Rank: 109/116
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_baseline
[sid:00070] GeoDesc-Upright w/ MAGSAC (n...
Size: 512 bytes. Matches: built-in
7829.63 545.49
Rank: 33/116
0.486
Rank: 36/116
0.848
Rank: 38/116
0.52078
(±0.00028)
Rank: 48/116
458.60
Rank: 55/116
4146.83
Rank: 48/116
4.412
Rank: 43/116
0.432
Rank: 52/116
0.70442
(±0.00189)
Rank: 42/116
0.61260
Rank: 43/116
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
[sid:00045] D2-Net (multi-scale), DEGENS...
Size: 2048 bytes. Matches: built-in
6924.14 345.93
Rank: 86/116
0.321
Rank: 110/116
0.386
Rank: 111/116
0.20486
(±0.00062)
Rank: 105/116
740.77
Rank: 18/116
5696.95
Rank: 9/116
3.326
Rank: 113/116
0.660
Rank: 109/116
0.42782
(±0.00544)
Rank: 109/116
0.31634
Rank: 104/116
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
[sid:00052] HardNet w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 371.67
Rank: 76/116
0.472
Rank: 78/116
0.853
Rank: 28/116
0.52325
(±0.00098)
Rank: 45/116
402.15
Rank: 71/116
3895.00
Rank: 58/116
4.345
Rank: 57/116
0.426
Rank: 46/116
0.69653
(±0.00108)
Rank: 48/116
0.60989
Rank: 47/116
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
[sid:00549] SIFT + DeepOrientation + SOS...
Size: 512 bytes. Matches: custom
2826.82 219.38
Rank: 106/116
0.284
Rank: 112/116
0.456
Rank: 109/116
0.14544
(±0.00000)
Rank: 108/116
221.40
Rank: 113/116
1839.87
Rank: 109/116
3.608
Rank: 106/116
0.641
Rank: 107/116
0.47073
(±0.00158)
Rank: 107/116
0.30808
Rank: 106/116
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
[sid:00113] VL-Hess-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 297.80
Rank: 95/116
0.547
Rank: 11/116
0.814
Rank: 83/116
0.44501
(±0.00077)
Rank: 88/116
347.39
Rank: 97/116
3209.10
Rank: 91/116
4.126
Rank: 91/116
0.519
Rank: 101/116
0.58657
(±0.00120)
Rank: 99/116
0.51579
Rank: 91/116
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
[sid:00567] Guided-HardNet32-v1-lib-qht-...
Size: 512 bytes. Matches: custom
7829.63 529.89
Rank: 37/116
0.486
Rank: 36/116
0.875
Rank: 6/116
0.58509
(±0.00000)
Rank: 5/116
536.93
Rank: 34/116
4685.16
Rank: 22/116
4.645
Rank: 6/116
0.376
Rank: 5/116
0.75702
(±0.00159)
Rank: 8/116
0.67105
Rank: 3/116
Anonymous (released after deadline) sift8k hardnet32-v1-lib-qht-p (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-15 is_baseline
[sid:00547] SIFT + DeepOrientation + SOS...
Size: 512 bytes. Matches: custom
2826.82 1096.89
Rank: 3/116
0.284
Rank: 112/116
0.166
Rank: 115/116
0.04936
(±0.00000)
Rank: 112/116
1106.68
Rank: 4/116
2494.00
Rank: 105/116
3.400
Rank: 110/116
0.646
Rank: 108/116
0.43054
(±0.00185)
Rank: 108/116
0.23995
Rank: 112/116
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
[sid:00537] SIFT + DeepOrientation + SOS...
Size: 512 bytes. Matches: custom
2826.82 1496.97
Rank: 1/116
0.284
Rank: 112/116
0.145
Rank: 116/116
0.00178
(±0.00000)
Rank: 116/116
1506.82
Rank: 1/116
2538.44
Rank: 104/116
3.388
Rank: 112/116
0.669
Rank: 110/116
0.42682
(±0.00259)
Rank: 110/116
0.21430
Rank: 113/116
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
[sid:00110] VL-DoGAff-SIFT-8k
Size: 512 bytes. Matches: built-in
7892.05 324.67
Rank: 91/116
0.482
Rank: 66/116
0.820
Rank: 79/116
0.46657
(±0.00079)
Rank: 80/116
311.54
Rank: 109/116
3061.54
Rank: 95/116
4.105
Rank: 95/116
0.475
Rank: 88/116
0.62964
(±0.00428)
Rank: 85/116
0.54810
Rank: 82/116
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
[sid:00072] L2-Net-Upright w/ DEGENSAC (...
Size: 512 bytes. Matches: built-in
7829.63 444.75
Rank: 56/116
0.486
Rank: 36/116
0.859
Rank: 23/116
0.54497
(±0.00003)
Rank: 26/116
395.53
Rank: 76/116
3603.85
Rank: 67/116
4.382
Rank: 49/116
0.452
Rank: 73/116
0.68491
(±0.00338)
Rank: 61/116
0.61494
Rank: 38/116
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
[sid:00590] Guided-HardNet-epoch4
Size: 512 bytes. Matches: custom
7829.63 596.85
Rank: 25/116
0.486
Rank: 36/116
0.875
Rank: 5/116
0.59919
(±0.00000)
Rank: 2/116
604.81
Rank: 24/116
5062.27
Rank: 14/116
4.710
Rank: 2/116
0.369
Rank: 3/116
0.76219
(±0.00253)
Rank: 3/116
0.68069
Rank: 2/116
Anonymous (released after deadline) sift8k hardnet-epoch4 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-24 is_baseline
[sid:00117] VL-HessAffNet-SIFT-8k
Size: 512 bytes. Matches: built-in
8000.00 306.56
Rank: 93/116
0.577
Rank: 5/116
0.831
Rank: 62/116
0.46793
(±0.00035)
Rank: 79/116
350.69
Rank: 92/116
3327.71
Rank: 86/116
4.076
Rank: 98/116
0.487
Rank: 91/116
0.60691
(±0.00338)
Rank: 93/116
0.53742
Rank: 86/116
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
[sid:00074] L2-Net w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 492.41
Rank: 44/116
0.472
Rank: 78/116
0.830
Rank: 68/116
0.52519
(±0.00039)
Rank: 41/116
339.00
Rank: 101/116
3424.94
Rank: 80/116
4.206
Rank: 83/116
0.448
Rank: 69/116
0.66445
(±0.00213)
Rank: 73/116
0.59482
Rank: 63/116
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
[sid:00595] SIFT8k_8000_HardNet64-train-...
Size: 512 bytes. Matches: built-in
7830.09 611.66
Rank: 22/116
0.486
Rank: 27/116
0.867
Rank: 15/116
0.58265
(±0.00102)
Rank: 8/116
786.40
Rank: 16/116
5595.19
Rank: 11/116
4.603
Rank: 14/116
0.390
Rank: 13/116
0.74372
(±0.00317)
Rank: 14/116
0.66319
Rank: 9/116
Anonymous (released after deadline) sift8k hardnet64-train-all-ben86-l2-210000 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-27 is_baseline
[sid:00521] SIFT8k-gift
Size: 512 bytes. Matches: built-in
6589.48 435.27
Rank: 59/116
0.467
Rank: 100/116
0.830
Rank: 64/116
0.48128
(±0.00056)
Rank: 74/116
452.18
Rank: 57/116
3770.17
Rank: 66/116
4.595
Rank: 17/116
0.444
Rank: 64/116
0.67443
(±0.00253)
Rank: 71/116
0.57785
Rank: 69/116
Anonymous (released after deadline) sift-def gift (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-26 is_submission
[sid:00566] Guided-HardNet2qhtp.json
Size: 512 bytes. Matches: custom
7829.63 527.15
Rank: 38/116
0.486
Rank: 36/116
0.874
Rank: 7/116
0.58251
(±0.00000)
Rank: 9/116
448.12
Rank: 59/116
4072.82
Rank: 50/116
4.607
Rank: 12/116
0.387
Rank: 11/116
0.74640
(±0.00246)
Rank: 12/116
0.66445
Rank: 7/116
Anonymous (released after deadline) sift8k hardnetnd32qhtp (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-15 is_baseline
[sid:00572] Guided HardNet qht
Size: 512 bytes. Matches: custom
7829.63 899.89
Rank: 7/116
0.486
Rank: 36/116
0.774
Rank: 94/116
0.54663
(±0.00000)
Rank: 25/116
538.28
Rank: 32/116
4687.01
Rank: 21/116
4.644
Rank: 7/116
0.376
Rank: 6/116
0.75721
(±0.00140)
Rank: 7/116
0.65192
Rank: 15/116
Anonymous (released after deadline) sift8k hardnet-qht (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-20 is_baseline
[sid:00541] sift and sosnet64 train scal...
Size: 512 bytes. Matches: built-in
7830.09 615.09
Rank: 21/116
0.486
Rank: 27/116
0.866
Rank: 16/116
0.58268
(±0.00061)
Rank: 7/116
957.95
Rank: 8/116
6312.38
Rank: 5/116
4.607
Rank: 13/116
0.385
Rank: 9/116
0.74476
(±0.00137)
Rank: 13/116
0.66372
Rank: 8/116
Anonymous (released after deadline) sift8k hardnet64-train-all-sos (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-01 is_submission
[sid:00102] KeyNet-HardNet-8k
Size: 512 bytes. Matches: built-in
7997.63 832.10
Rank: 9/116
0.582
Rank: 1/116
0.788
Rank: 89/116
0.47395
(±0.00073)
Rank: 75/116
356.21
Rank: 88/116
3366.01
Rank: 83/116
4.319
Rank: 64/116
0.464
Rank: 82/116
0.64829
(±0.00174)
Rank: 81/116
0.56112
Rank: 80/116
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
[sid:00556] HarrisZ + DeepOrientation + ...
Size: 512 bytes. Matches: custom
2410.18 390.12
Rank: 69/116
0.368
Rank: 105/116
0.646
Rank: 104/116
0.04192
(±0.00000)
Rank: 113/116
392.52
Rank: 82/116
1475.19
Rank: 113/116
4.454
Rank: 41/116
0.601
Rank: 105/116
0.48498
(±0.00118)
Rank: 105/116
0.26345
Rank: 111/116
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_baseline
[sid:00073] L2-Net-Upright w/ MAGSAC (no...
Size: 512 bytes. Matches: built-in
7829.63 590.66
Rank: 27/116
0.486
Rank: 36/116
0.837
Rank: 57/116
0.53911
(±0.00062)
Rank: 31/116
395.53
Rank: 76/116
3603.85
Rank: 67/116
4.382
Rank: 49/116
0.455
Rank: 75/116
0.68491
(±0.00338)
Rank: 61/116
0.61201
Rank: 44/116
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
[sid:00080] LogPolar w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 384.53
Rank: 72/116
0.472
Rank: 78/116
0.837
Rank: 55/116
0.50188
(±0.00044)
Rank: 62/116
399.70
Rank: 73/116
4036.37
Rank: 53/116
4.340
Rank: 60/116
0.429
Rank: 48/116
0.69100
(±0.00154)
Rank: 54/116
0.59644
Rank: 61/116
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
[sid:00006] Upright SIFT (OpenCV), DEGEN...
Size: 512 bytes. Matches: built-in
7829.24 326.21
Rank: 90/116
0.487
Rank: 21/116
0.830
Rank: 66/116
0.48742
(±0.00042)
Rank: 72/116
525.57
Rank: 36/116
4147.20
Rank: 47/116
4.258
Rank: 77/116
0.460
Rank: 78/116
0.65664
(±0.00231)
Rank: 77/116
0.57203
Rank: 75/116
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
[sid:00059] L2-Net w/ DEGENSAC
Size: 512 bytes. Matches: built-in
7861.11 304.61
Rank: 94/116
0.472
Rank: 78/116
0.835
Rank: 60/116
0.49523
(±0.00042)
Rank: 66/116
314.54
Rank: 107/116
3314.19
Rank: 89/116
4.204
Rank: 85/116
0.467
Rank: 85/116
0.65665
(±0.00287)
Rank: 75/116
0.57594
Rank: 71/116
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
[sid:00050] GeoDesc-Upright w/ MAGSAC
Size: 512 bytes. Matches: built-in
7829.63 419.68
Rank: 62/116
0.486
Rank: 36/116
0.827
Rank: 70/116
0.48259
(±0.00060)
Rank: 73/116
394.67
Rank: 80/116
3863.42
Rank: 60/116
4.380
Rank: 51/116
0.443
Rank: 61/116
0.67682
(±0.00127)
Rank: 69/116
0.57971
Rank: 68/116
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
[sid:00054] HardNet w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 488.84
Rank: 45/116
0.472
Rank: 78/116
0.830
Rank: 67/116
0.51809
(±0.00053)
Rank: 51/116
402.15
Rank: 71/116
3895.00
Rank: 58/116
4.345
Rank: 57/116
0.423
Rank: 43/116
0.69653
(±0.00108)
Rank: 48/116
0.60731
Rank: 52/116
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
[sid:00067] GeoDesc w/ MAGSAC (no FLANN)
Size: 512 bytes. Matches: built-in
7861.11 463.81
Rank: 51/116
0.472
Rank: 78/116
0.835
Rank: 58/116
0.50559
(±0.00081)
Rank: 60/116
395.12
Rank: 78/116
3838.97
Rank: 62/116
4.264
Rank: 75/116
0.442
Rank: 59/116
0.68032
(±0.00099)
Rank: 67/116
0.59295
Rank: 64/116
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
[sid:00540] ASLV2+OANetV2+DEGENSAC
Size: 512 bytes. Matches: custom
5982.87 1053.74
Rank: 4/116
0.550
Rank: 10/116
0.758
Rank: 96/116
0.50173
(±0.00000)
Rank: 63/116
1067.11
Rank: 5/116
4587.47
Rank: 25/116
4.948
Rank: 1/116
0.386
Rank: 10/116
0.75275
(±0.00186)
Rank: 9/116
0.62724
Rank: 31/116
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
[sid:00514] UprightRootSIFT-AdaLAM
Size: 512 bytes. Matches: custom
6449.42 632.63
Rank: 15/116
0.436
Rank: 102/116
0.758
Rank: 97/116
0.31372
(±0.00000)
Rank: 99/116
640.08
Rank: 21/116
5286.27
Rank: 13/116
4.343
Rank: 59/116
0.398
Rank: 16/116
0.72130
(±0.00085)
Rank: 27/116
0.51751
Rank: 90/116
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. N/A N/A 20-04-25 is_submission
[sid:00511] r2d2-wasfi
Size: 512 bytes. Matches: built-in
7861.03 281.30
Rank: 97/116
0.503
Rank: 16/116
0.762
Rank: 95/116
0.37893
(±0.00023)
Rank: 95/116
416.35
Rank: 67/116
2834.69
Rank: 101/116
4.293
Rank: 71/116
0.480
Rank: 90/116
0.64359
(±0.00103)
Rank: 84/116
0.51126
Rank: 94/116
Anonymous (released after deadline) r2d2 r2d2-wasfi-epoch25-pretrained-n16-8k (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00083] LogPolar-Upright w/ DEGENSAC...
Size: 512 bytes. Matches: built-in
7829.63 553.21
Rank: 31/116
0.486
Rank: 36/116
0.865
Rank: 19/116
0.55102
(±0.00015)
Rank: 23/116
505.37
Rank: 48/116
4414.11
Rank: 34/116
4.518
Rank: 33/116
0.422
Rank: 40/116
0.71092
(±0.00251)
Rank: 33/116
0.63097
Rank: 27/116
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
[sid:00064] SOSNet w/ MAGSAC
Size: 512 bytes. Matches: built-in
7861.11 455.02
Rank: 53/116
0.472
Rank: 78/116
0.831
Rank: 63/116
0.51402
(±0.00057)
Rank: 54/116
440.21
Rank: 60/116
4178.82
Rank: 43/116
4.378
Rank: 53/116
0.422
Rank: 41/116
0.70078
(±0.00375)
Rank: 45/116
0.60740
Rank: 51/116
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
[sid:00104] VL-DoG-SIFT-8k
Size: 512 bytes. Matches: built-in
7880.59 183.65
Rank: 109/116
0.490
Rank: 18/116
0.835
Rank: 59/116
0.39987
(±0.00068)
Rank: 91/116
324.62
Rank: 104/116
3030.67
Rank: 98/116
4.173
Rank: 87/116
0.467
Rank: 86/116
0.62829
(±0.00112)
Rank: 89/116
0.51408
Rank: 92/116
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
[sid:00555] SuperPoint+GIFT+Graph Motion...
Size: 512 bytes. Matches: custom
1940.85 286.50
Rank: 3/61
0.356
Rank: 18/61
0.676
Rank: 45/61
0.42283
(±0.00000)
Rank: 15/61
290.06
Rank: 15/61
1887.70
Rank: 9/61
4.567
Rank: 7/61
0.420
Rank: 3/61
0.70395
(±0.00239)
Rank: 3/61
0.56339
Rank: 5/61
Anonymous (released after deadline) superpoint-2k scale-gift (128 float32: 512 bytes) Anonymous (released after deadline) N/A Anonymous (released after deadline) 20-05-10 is_baseline
[sid:00038] SOSNet, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 137.72
Rank: 26/61
0.323
Rank: 37/61
0.760
Rank: 34/61
0.38654
(±0.00056)
Rank: 27/61
145.47
Rank: 53/61
1271.41
Rank: 38/61
4.045
Rank: 37/61
0.526
Rank: 34/61
0.56072
(±0.00077)
Rank: 34/61
0.47363
Rank: 30/61
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
[sid:00007] L2-Net, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 89.22
Rank: 54/61
0.323
Rank: 43/61
0.779
Rank: 23/61
0.36265
(±0.00090)
Rank: 35/61
138.97
Rank: 55/61
1211.49
Rank: 43/61
3.948
Rank: 45/61
0.537
Rank: 37/61
0.52019
(±0.00120)
Rank: 42/61
0.44142
Rank: 40/61
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
[sid:00516] Key.Net (MS 4-1, Pyramid) + ...
Size: 512 bytes. Matches: built-in
2039.11 271.97
Rank: 4/61
0.447
Rank: 4/61
0.776
Rank: 26/61
0.40559
(±0.00112)
Rank: 18/61
336.81
Rank: 10/61
1559.74
Rank: 18/61
4.536
Rank: 8/61
0.461
Rank: 11/61
0.66406
(±0.00521)
Rank: 8/61
0.53483
Rank: 12/61
Anonymous (released after deadline) keynet-41-pyramidv2 x-net-lib (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-24 is_submission
[sid:00002] SURF (OpenCV), DEGENSAC
Size: 256 bytes. Matches: built-in
2010.88 73.55
Rank: 57/61
0.321
Rank: 49/61
0.634
Rank: 46/61
0.20761
(±0.00029)
Rank: 50/61
300.02
Rank: 14/61
1208.06
Rank: 45/61
3.549
Rank: 50/61
0.695
Rank: 50/61
0.37845
(±0.00399)
Rank: 50/61
0.29303
Rank: 50/61
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
[sid:00544] Key.Net + X-Net(Lib) w/ DEGE...
Size: 512 bytes. Matches: built-in
2039.11 198.77
Rank: 6/61
0.447
Rank: 4/61
0.814
Rank: 6/61
0.42332
(±0.00096)
Rank: 14/61
312.61
Rank: 12/61
1515.67
Rank: 19/61
4.567
Rank: 6/61
0.448
Rank: 5/61
0.66559
(±0.00233)
Rank: 7/61
0.54445
Rank: 6/61
Anonymous (released after deadline) keynet x-net-lib (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-28 is_submission
[sid:00523] SEKD
Size: 512 bytes. Matches: built-in
2043.44 119.88
Rank: 33/61
0.386
Rank: 10/61
0.799
Rank: 13/61
0.42811
(±0.00096)
Rank: 12/61
166.41
Rank: 41/61
1163.31
Rank: 48/61
4.427
Rank: 11/61
0.456
Rank: 8/61
0.64919
(±0.00171)
Rank: 9/61
0.53865
Rank: 9/61
Anonymous (released after deadline) sekd sekd (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-29 is_submission
[sid:00575] SIFT2k-2048-HardNet64-rain-a...
Size: 512 bytes. Matches: built-in
2047.77 165.12
Rank: 12/61
0.339
Rank: 20/61
0.814
Rank: 4/61
0.43543
(±0.00051)
Rank: 8/61
255.45
Rank: 20/61
1708.29
Rank: 13/61
4.327
Rank: 15/61
0.451
Rank: 7/61
0.64653
(±0.00221)
Rank: 10/61
0.54098
Rank: 7/61
Anonymous (released after deadline) sift2k sift2k-2048-hardnet64-train-all-sos-812000 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-20 is_baseline
[sid:00041] SuperPoint (2k features, NMS...
Size: 1024 bytes. Matches: built-in
2048.00 120.09
Rank: 32/61
0.368
Rank: 13/61
0.628
Rank: 48/61
0.28679
(±0.00056)
Rank: 47/61
156.04
Rank: 45/61
1121.20
Rank: 52/61
4.310
Rank: 19/61
0.566
Rank: 45/61
0.53829
(±0.00021)
Rank: 39/61
0.41254
Rank: 45/61
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
[sid:00001] HardNet, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 112.03
Rank: 41/61
0.323
Rank: 37/61
0.776
Rank: 25/61
0.38578
(±0.00131)
Rank: 29/61
153.29
Rank: 48/61
1306.64
Rank: 34/61
4.026
Rank: 39/61
0.539
Rank: 40/61
0.55573
(±0.00175)
Rank: 36/61
0.47075
Rank: 32/61
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
[sid:00565] SIFT2k_2000_HardNet64-train-...
Size: 512 bytes. Matches: built-in
1999.83 160.96
Rank: 14/61
0.336
Rank: 21/61
0.813
Rank: 7/61
0.43188
(±0.00005)
Rank: 9/61
248.91
Rank: 22/61
1666.27
Rank: 15/61
4.323
Rank: 16/61
0.465
Rank: 14/61
0.64404
(±0.00271)
Rank: 12/61
0.53796
Rank: 10/61
Anonymous (released after deadline) sift2k sift2k-2000-hardnet64-train-all-sos-812000 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-13 is_baseline
[sid:00011] GeoDesc, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 95.61
Rank: 51/61
0.323
Rank: 43/61
0.759
Rank: 35/61
0.34215
(±0.00054)
Rank: 39/61
120.68
Rank: 59/61
1098.70
Rank: 53/61
4.002
Rank: 42/61
0.552
Rank: 42/61
0.50594
(±0.00073)
Rank: 45/61
0.42405
Rank: 42/61
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
[sid:00043] DELF-GLD (128D), PyRANSAC
Size: 512 bytes. Matches: built-in
2036.82 96.75
Rank: 50/61
0.109
Rank: 55/61
0.156
Rank: 57/61
0.07298
(±0.00056)
Rank: 59/61
430.68
Rank: 5/61
2162.72
Rank: 3/61
2.535
Rank: 57/61
0.887
Rank: 58/61
0.13247
(±0.00116)
Rank: 56/61
0.10272
Rank: 57/61
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
[sid:00507] sift and hardnet512
Size: 2048 bytes. Matches: built-in
2000.00 100.16
Rank: 48/61
0.321
Rank: 51/61
0.778
Rank: 24/61
0.35047
(±0.00046)
Rank: 38/61
153.92
Rank: 47/61
1310.53
Rank: 33/61
4.017
Rank: 41/61
0.515
Rank: 28/61
0.56598
(±0.00185)
Rank: 32/61
0.45822
Rank: 37/61
Anonymous (released after deadline) sift8k hardnet512 (512 float32: 2048 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00578] Key.Net-s + DescNet w/ DEGEN...
Size: 512 bytes. Matches: built-in
2033.67 252.57
Rank: 5/61
0.449
Rank: 2/61
0.805
Rank: 10/61
0.45418
(±0.00056)
Rank: 4/61
331.55
Rank: 11/61
1621.68
Rank: 17/61
4.570
Rank: 5/61
0.447
Rank: 4/61
0.67414
(±0.00371)
Rank: 5/61
0.56416
Rank: 4/61
Anonymous (released after deadline) keynet-s descnet (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-20 is_baseline
[sid:00030] Upright Root-SIFT (OpenCV), ...
Size: 512 bytes. Matches: built-in
1892.72 114.22
Rank: 39/61
0.333
Rank: 23/61
0.781
Rank: 22/61
0.39547
(±0.00027)
Rank: 22/61
201.15
Rank: 27/61
1347.75
Rank: 29/61
4.091
Rank: 33/61
0.522
Rank: 31/61
0.56078
(±0.00264)
Rank: 33/61
0.47812
Rank: 27/61
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
[sid:00009] GeoDesc (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 119.25
Rank: 35/61
0.333
Rank: 25/61
0.784
Rank: 19/61
0.38660
(±0.00078)
Rank: 26/61
149.66
Rank: 51/61
1237.96
Rank: 40/61
4.219
Rank: 28/61
0.525
Rank: 33/61
0.56687
(±0.00261)
Rank: 30/61
0.47673
Rank: 28/61
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
[sid:00017] ORB (OpenCV), DEGENSAC
Size: 32 bytes. Matches: built-in
2031.83 64.72
Rank: 60/61
0.366
Rank: 14/61
0.523
Rank: 52/61
0.08243
(±0.00058)
Rank: 56/61
120.93
Rank: 58/61
280.38
Rank: 61/61
2.191
Rank: 61/61
0.863
Rank: 54/61
0.02318
(±0.00229)
Rank: 61/61
0.05280
Rank: 61/61
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
[sid:00028] DELF-GLD (512D), DEGENSAC
Size: 2048 bytes. Matches: built-in
2036.82 101.15
Rank: 47/61
0.109
Rank: 55/61
0.170
Rank: 56/61
0.09156
(±0.00045)
Rank: 54/61
437.64
Rank: 3/61
2296.72
Rank: 1/61
2.567
Rank: 55/61
0.877
Rank: 55/61
0.14991
(±0.00513)
Rank: 54/61
0.12073
Rank: 54/61
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
[sid:00589] SuperPoint-128d + SuperGlue ...
Size: 512 bytes. Matches: custom
1973.62 325.80
Rank: 2/61
0.364
Rank: 16/61
0.772
Rank: 28/61
0.55214
(±0.00098)
Rank: 2/61
429.49
Rank: 7/61
2130.73
Rank: 6/61
4.570
Rank: 4/61
0.384
Rank: 2/61
0.75283
(±0.00086)
Rank: 1/61
0.65248
Rank: 1/61
Anonymous (released after deadline) superpoint-k2048-nms3-refine2-r1600forcecubic superpoint-down128 (128 float32: 512 bytes) Anonymous (released after deadline) https://arxiv.org/abs/1911.11763 Anonymous (released after deadline) 20-05-24 is_baseline
[sid:00029] SuperPoint (2k features, NMS...
Size: 1024 bytes. Matches: built-in
2048.00 125.75
Rank: 29/61
0.364
Rank: 17/61
0.630
Rank: 47/61
0.28965
(±0.00063)
Rank: 46/61
170.60
Rank: 38/61
1185.38
Rank: 47/61
4.331
Rank: 14/61
0.557
Rank: 43/61
0.54663
(±0.00218)
Rank: 38/61
0.41814
Rank: 44/61
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
[sid:00046] D2-Net (single-scale), DEGEN...
Size: 2048 bytes. Matches: built-in
2038.22 151.10
Rank: 21/61
0.202
Rank: 52/61
0.337
Rank: 53/61
0.16324
(±0.00082)
Rank: 52/61
364.68
Rank: 9/61
1984.98
Rank: 8/61
3.412
Rank: 51/61
0.720
Rank: 51/61
0.36911
(±0.00167)
Rank: 51/61
0.26618
Rank: 51/61
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
[sid:00036] SOSNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 159.24
Rank: 16/61
0.333
Rank: 25/61
0.794
Rank: 16/61
0.43077
(±0.00017)
Rank: 10/61
183.92
Rank: 31/61
1403.66
Rank: 24/61
4.314
Rank: 17/61
0.472
Rank: 17/61
0.62181
(±0.00161)
Rank: 20/61
0.52629
Rank: 18/61
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
[sid:00012] GeoDesc (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 148.30
Rank: 22/61
0.333
Rank: 25/61
0.765
Rank: 32/61
0.38124
(±0.00056)
Rank: 32/61
149.66
Rank: 51/61
1237.96
Rank: 40/61
4.219
Rank: 28/61
0.526
Rank: 35/61
0.56687
(±0.00261)
Rank: 30/61
0.47405
Rank: 29/61
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
[sid:00026] AKAZE (OpenCV), DEGENSAC
Size: 61 bytes. Matches: built-in
1657.83 70.81
Rank: 59/61
0.355
Rank: 19/61
0.622
Rank: 50/61
0.17632
(±0.00087)
Rank: 51/61
224.40
Rank: 24/61
795.54
Rank: 60/61
3.311
Rank: 52/61
0.819
Rank: 53/61
0.23020
(±0.00466)
Rank: 53/61
0.20326
Rank: 52/61
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
[sid:00583] R2D2
Size: 512 bytes. Matches: built-in
2048.00 172.33
Rank: 11/61
0.398
Rank: 9/61
0.695
Rank: 44/61
0.33715
(±0.00149)
Rank: 41/61
306.15
Rank: 13/61
1235.83
Rank: 42/61
4.156
Rank: 32/61
0.496
Rank: 23/61
0.60223
(±0.00103)
Rank: 25/61
0.46969
Rank: 34/61
(contact) r2d2-5k-p-aug r2d2-5k-p-aug (128 float32: 512 bytes) N/A N/A 20-05-23 is_baseline
[sid:00513] Personal Inplementation SIFT...
Size: 512 bytes. Matches: built-in
1900.60 70.91
Rank: 58/61
0.085
Rank: 61/61
0.002
Rank: 61/61
0.07782
(±0.00022)
Rank: 57/61
173.06
Rank: 34/61
948.78
Rank: 59/61
3.816
Rank: 47/61
0.887
Rank: 59/61
0.09835
(±0.00068)
Rank: 58/61
0.08809
Rank: 58/61
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
[sid:00013] SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1936.30 78.65
Rank: 56/61
0.323
Rank: 47/61
0.733
Rank: 41/61
0.30766
(±0.00075)
Rank: 45/61
171.92
Rank: 37/61
1059.70
Rank: 57/61
3.687
Rank: 49/61
0.618
Rank: 48/61
0.42476
(±0.00103)
Rank: 48/61
0.36621
Rank: 48/61
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
[sid:00123] KeyNet-HardNet-2k
Size: 512 bytes. Matches: built-in
2048.00 137.55
Rank: 27/61
0.469
Rank: 1/61
0.818
Rank: 3/61
0.32725
(±0.00026)
Rank: 43/61
195.33
Rank: 28/61
1276.30
Rank: 37/61
4.493
Rank: 9/61
0.490
Rank: 22/61
0.61606
(±0.00141)
Rank: 22/61
0.47166
Rank: 31/61
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
[sid:00044] Upright SIFT (OpenCV), DEGEN...
Size: 512 bytes. Matches: built-in
1892.72 106.58
Rank: 45/61
0.333
Rank: 23/61
0.763
Rank: 33/61
0.37078
(±0.00056)
Rank: 34/61
205.17
Rank: 26/61
1300.49
Rank: 36/61
4.000
Rank: 44/61
0.563
Rank: 44/61
0.52549
(±0.00120)
Rank: 41/61
0.44814
Rank: 39/61
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
[sid:00503] Guided matching hardnet
Size: 512 bytes. Matches: custom
1892.71 152.91
Rank: 19/61
0.333
Rank: 34/61
0.795
Rank: 15/61
0.42159
(±0.00000)
Rank: 16/61
254.13
Rank: 21/61
1754.99
Rank: 12/61
4.283
Rank: 24/61
0.464
Rank: 13/61
0.64144
(±0.00207)
Rank: 14/61
0.53152
Rank: 14/61
Anonymous (released after deadline) sift upright-hardnet (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
[sid:00022] L2-Net (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 113.44
Rank: 40/61
0.333
Rank: 25/61
0.801
Rank: 12/61
0.40729
(±0.00097)
Rank: 17/61
171.93
Rank: 35/61
1333.56
Rank: 30/61
4.226
Rank: 26/61
0.497
Rank: 25/61
0.58762
(±0.00228)
Rank: 26/61
0.49746
Rank: 22/61
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
[sid:00016] LogPolarDesc, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 151.94
Rank: 20/61
0.323
Rank: 37/61
0.752
Rank: 39/61
0.38840
(±0.00064)
Rank: 25/61
162.84
Rank: 43/61
1385.60
Rank: 27/61
4.046
Rank: 35/61
0.516
Rank: 29/61
0.57149
(±0.00344)
Rank: 28/61
0.47994
Rank: 26/61
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
[sid:00032] Root-SIFT (OpenCV), DEGENSAC
Size: 512 bytes. Matches: built-in
1936.30 85.35
Rank: 55/61
0.323
Rank: 47/61
0.756
Rank: 38/61
0.33605
(±0.00092)
Rank: 42/61
167.84
Rank: 40/61
1141.88
Rank: 50/61
3.788
Rank: 48/61
0.576
Rank: 46/61
0.46826
(±0.00154)
Rank: 47/61
0.40216
Rank: 46/61
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
[sid:00040] DELF-GLD (128D), DEGENSAC
Size: 512 bytes. Matches: built-in
2036.82 103.04
Rank: 46/61
0.109
Rank: 55/61
0.151
Rank: 58/61
0.08367
(±0.00047)
Rank: 55/61
430.68
Rank: 5/61
2162.72
Rank: 3/61
2.535
Rank: 57/61
0.880
Rank: 57/61
0.13247
(±0.00116)
Rank: 56/61
0.10807
Rank: 56/61
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
[sid:00515] SuperPoint + SuperGlue + DEG...
Size: 1024 bytes. Matches: custom
1973.62 325.83
Rank: 1/61
0.364
Rank: 15/61
0.772
Rank: 27/61
0.55234
(±0.00024)
Rank: 1/61
429.48
Rank: 8/61
2131.07
Rank: 5/61
4.572
Rank: 3/61
0.373
Rank: 1/61
0.75213
(±0.00345)
Rank: 2/61
0.65224
Rank: 2/61
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
[sid:00027] DELF-GLD (32D), PyRANSAC
Size: 128 bytes. Matches: built-in
2036.82 92.99
Rank: 53/61
0.109
Rank: 55/61
0.122
Rank: 59/61
0.06197
(±0.00032)
Rank: 61/61
453.19
Rank: 1/61
1805.48
Rank: 10/61
2.425
Rank: 59/61
0.916
Rank: 60/61
0.09163
(±0.00140)
Rank: 59/61
0.07680
Rank: 60/61
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
[sid:00037] GeoDesc, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 117.86
Rank: 37/61
0.323
Rank: 43/61
0.742
Rank: 40/61
0.34024
(±0.00050)
Rank: 40/61
120.68
Rank: 59/61
1098.70
Rank: 53/61
4.002
Rank: 42/61
0.551
Rank: 41/61
0.50594
(±0.00073)
Rank: 45/61
0.42309
Rank: 43/61
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
[sid:00023] LogPolarDesc, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 119.51
Rank: 34/61
0.323
Rank: 37/61
0.770
Rank: 29/61
0.39335
(±0.00060)
Rank: 23/61
162.84
Rank: 43/61
1385.60
Rank: 27/61
4.046
Rank: 35/61
0.519
Rank: 30/61
0.57149
(±0.00344)
Rank: 28/61
0.48242
Rank: 24/61
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
[sid:00564] SIFT2k_2000_HardNet64_train_...
Size: 512 bytes. Matches: built-in
1999.83 160.13
Rank: 15/61
0.336
Rank: 21/61
0.814
Rank: 5/61
0.42959
(±0.00060)
Rank: 11/61
244.13
Rank: 23/61
1646.76
Rank: 16/61
4.334
Rank: 13/61
0.459
Rank: 10/61
0.64407
(±0.00339)
Rank: 11/61
0.53683
Rank: 11/61
Anonymous (released after deadline) sift2k sift2k-2000-hardnet64-train-all-l2-138000 (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-13 is_baseline
[sid:00025] DELF-GLD (32D), DEGENSAC
Size: 128 bytes. Matches: built-in
2036.82 98.57
Rank: 49/61
0.109
Rank: 55/61
0.119
Rank: 60/61
0.06573
(±0.00030)
Rank: 60/61
453.19
Rank: 1/61
1805.48
Rank: 10/61
2.425
Rank: 59/61
0.920
Rank: 61/61
0.09163
(±0.00140)
Rank: 59/61
0.07868
Rank: 59/61
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
[sid:00512] Hardnet with DEGENSAC for st...
Size: 512 bytes. Matches: built-in
1999.50 52.95
Rank: 61/61
0.172
Rank: 53/61
0.558
Rank: 51/61
0.20771
(±0.00049)
Rank: 49/61
181.48
Rank: 33/61
1692.63
Rank: 14/61
3.050
Rank: 53/61
0.695
Rank: 49/61
0.38463
(±0.00275)
Rank: 49/61
0.29617
Rank: 49/61
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
[sid:00527] SEKD
Size: 512 bytes. Matches: built-in
1786.69 108.69
Rank: 43/61
0.383
Rank: 11/61
0.759
Rank: 36/61
0.38886
(±0.00013)
Rank: 24/61
150.11
Rank: 50/61
1063.08
Rank: 56/61
4.575
Rank: 2/61
0.465
Rank: 15/61
0.63303
(±0.00244)
Rank: 15/61
0.51095
Rank: 21/61
Anonymous (released after deadline) sekd sekd (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-27 is_submission
[sid:00031] SuperPoint (2k features, NMS...
Size: 1024 bytes. Matches: built-in
2048.00 108.26
Rank: 44/61
0.376
Rank: 12/61
0.624
Rank: 49/61
0.27954
(±0.00073)
Rank: 48/61
130.52
Rank: 57/61
1002.03
Rank: 58/61
4.211
Rank: 30/61
0.593
Rank: 47/61
0.51061
(±0.00378)
Rank: 44/61
0.39508
Rank: 47/61
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
[sid:00021] DELF-GLD (512D), PyRANSAC
Size: 2048 bytes. Matches: built-in
2036.82 95.04
Rank: 52/61
0.109
Rank: 55/61
0.175
Rank: 55/61
0.07687
(±0.00010)
Rank: 58/61
437.64
Rank: 3/61
2296.72
Rank: 1/61
2.567
Rank: 55/61
0.878
Rank: 56/61
0.14991
(±0.00513)
Rank: 54/61
0.11339
Rank: 55/61
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
[sid:00586] KeyNet+HardNet+NM-Net_v2
Size: 512 bytes. Matches: custom
2042.89 163.56
Rank: 13/61
0.435
Rank: 6/61
0.821
Rank: 1/61
0.40264
(±0.00000)
Rank: 19/61
165.01
Rank: 42/61
1134.69
Rank: 51/61
4.720
Rank: 1/61
0.448
Rank: 6/61
0.67488
(±0.00197)
Rank: 4/61
0.53876
Rank: 8/61
Anonymous (released after deadline) keynet hardnet (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-23 is_baseline
[sid:00020] D2-Net (multi-scale), DEGENS...
Size: 2048 bytes. Matches: built-in
2045.55 121.15
Rank: 31/61
0.168
Rank: 54/61
0.293
Rank: 54/61
0.12270
(±0.00048)
Rank: 53/61
286.10
Rank: 16/61
1999.37
Rank: 7/61
3.013
Rank: 54/61
0.774
Rank: 52/61
0.28297
(±0.00254)
Rank: 52/61
0.20283
Rank: 53/61
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
[sid:00585] NM-Net_v2
Size: 512 bytes. Matches: custom
1892.70 154.75
Rank: 18/61
0.333
Rank: 35/61
0.809
Rank: 8/61
0.46620
(±0.00000)
Rank: 3/61
155.91
Rank: 46/61
1392.47
Rank: 26/61
4.424
Rank: 12/61
0.459
Rank: 9/61
0.66837
(±0.00125)
Rank: 6/61
0.56729
Rank: 3/61
Anonymous (released after deadline) siftdef hardnet (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-05-23 is_baseline
[sid:00517] retrained Hardnet with MAGSA...
Size: 512 bytes. Matches: built-in
1892.71 136.55
Rank: 28/61
0.333
Rank: 25/61
0.808
Rank: 9/61
0.40042
(±0.00077)
Rank: 20/61
210.10
Rank: 25/61
1480.67
Rank: 20/61
4.226
Rank: 25/61
0.479
Rank: 21/61
0.63030
(±0.00242)
Rank: 16/61
0.51536
Rank: 19/61
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
[sid:00509] Upright-Sift + X-Net-lib w/ ...
Size: 512 bytes. Matches: built-in
1892.70 119.25
Rank: 36/61
0.333
Rank: 36/61
0.821
Rank: 2/61
0.42382
(±0.00089)
Rank: 13/61
168.67
Rank: 39/61
1319.63
Rank: 32/61
4.303
Rank: 20/61
0.468
Rank: 16/61
0.62953
(±0.00216)
Rank: 17/61
0.52668
Rank: 17/61
Anonymous (released after deadline) sift-def x-net-lib-upright-no-dups (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-25 is_submission
CV-DoG-HardNet8-PT
Size: 512 bytes. Matches: built-in
2048.00 117.11
Rank: 38/61
0.321
Rank: 50/61
0.790
Rank: 17/61
0.38200
(±0.00081)
Rank: 31/61
117.84
Rank: 61/61
1067.13
Rank: 55/61
4.051
Rank: 34/61
0.531
Rank: 36/61
0.53035
(±0.00531)
Rank: 40/61
0.45617
Rank: 38/61
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_baseline
[sid:00003] HardNet (upright), DEGENSAC
Size: 512 bytes. Matches: built-in
1892.71 144.79
Rank: 23/61
0.333
Rank: 25/61
0.803
Rank: 11/61
0.44235
(±0.00127)
Rank: 5/61
193.09
Rank: 29/61
1435.93
Rank: 22/61
4.302
Rank: 21/61
0.474
Rank: 19/61
0.62275
(±0.00423)
Rank: 18/61
0.53255
Rank: 13/61
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
[sid:00035] HardNet (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 185.87
Rank: 7/61
0.333
Rank: 25/61
0.781
Rank: 21/61
0.43816
(±0.00058)
Rank: 6/61
193.09
Rank: 29/61
1435.93
Rank: 22/61
4.302
Rank: 21/61
0.473
Rank: 18/61
0.62275
(±0.00423)
Rank: 18/61
0.53045
Rank: 15/61
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
[sid:00004] SOSNet (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 180.53
Rank: 8/61
0.333
Rank: 25/61
0.782
Rank: 20/61
0.43803
(±0.00092)
Rank: 7/61
183.92
Rank: 31/61
1403.66
Rank: 24/61
4.314
Rank: 17/61
0.474
Rank: 20/61
0.62181
(±0.00161)
Rank: 20/61
0.52992
Rank: 16/61
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
[sid:00018] R2D2 (r2d2-wasf-n16), DEGENS...
Size: 512 bytes. Matches: built-in
2048.00 179.83
Rank: 9/61
0.422
Rank: 7/61
0.725
Rank: 42/61
0.35783
(±0.00061)
Rank: 36/61
262.02
Rank: 19/61
1188.11
Rank: 46/61
4.299
Rank: 23/61
0.499
Rank: 26/61
0.60639
(±0.00124)
Rank: 23/61
0.48211
Rank: 25/61
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
[sid:00014] L2-Net, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 110.16
Rank: 42/61
0.323
Rank: 43/61
0.766
Rank: 31/61
0.35649
(±0.00110)
Rank: 37/61
138.97
Rank: 55/61
1211.49
Rank: 43/61
3.948
Rank: 45/61
0.538
Rank: 39/61
0.52019
(±0.00120)
Rank: 42/61
0.43834
Rank: 41/61
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
[sid:00033] SOSNet, DEGENSAC
Size: 512 bytes. Matches: built-in
1936.28 122.67
Rank: 30/61
0.323
Rank: 37/61
0.769
Rank: 30/61
0.37632
(±0.00056)
Rank: 33/61
145.47
Rank: 53/61
1271.41
Rank: 38/61
4.045
Rank: 37/61
0.524
Rank: 32/61
0.56072
(±0.00077)
Rank: 34/61
0.46852
Rank: 35/61
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
[sid:00039] HardNet, MAGSAC
Size: 512 bytes. Matches: built-in
1936.28 141.63
Rank: 25/61
0.323
Rank: 37/61
0.757
Rank: 37/61
0.38490
(±0.00120)
Rank: 30/61
153.29
Rank: 48/61
1306.64
Rank: 34/61
4.026
Rank: 39/61
0.537
Rank: 38/61
0.55573
(±0.00175)
Rank: 36/61
0.47032
Rank: 33/61
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
[sid:00582] R2D2
Size: 512 bytes. Matches: built-in
2048.00 157.87
Rank: 17/61
0.403
Rank: 8/61
0.709
Rank: 43/61
0.32501
(±0.00085)
Rank: 44/61
285.11
Rank: 17/61
1162.94
Rank: 49/61
4.161
Rank: 31/61
0.508
Rank: 27/61
0.60276
(±0.00062)
Rank: 24/61
0.46389
Rank: 36/61
(contact) r2d2-5k-p r2d2-5k-p (128 float32: 512 bytes) N/A N/A 20-05-23 is_baseline
[sid:00008] L2-Net (upright), MAGSAC
Size: 512 bytes. Matches: built-in
1892.71 141.92
Rank: 24/61
0.333
Rank: 25/61
0.786
Rank: 18/61
0.39980
(±0.00020)
Rank: 21/61
171.93
Rank: 35/61
1333.56
Rank: 30/61
4.226
Rank: 26/61
0.496
Rank: 24/61
0.58762
(±0.00228)
Rank: 26/61
0.49371
Rank: 23/61
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
[sid:00531] Key.Net + X-Net w/ DEGENSAC
Size: 512 bytes. Matches: built-in
2040.08 177.80
Rank: 10/61
0.447
Rank: 3/61
0.797
Rank: 14/61
0.38608
(±0.00068)
Rank: 28/61
276.91
Rank: 18/61
1468.50
Rank: 21/61
4.447
Rank: 10/61
0.463
Rank: 12/61
0.64369
(±0.00081)
Rank: 13/61
0.51489
Rank: 20/61
Anonymous (released after deadline) keynet x-net (128 float32: 512 bytes) Anonymous (released after deadline) N/A N/A 20-04-24 is_submission