Submission ID: 00603
SuperPoint-128d-masked + SuperGlue + DEGENSAC
Processed: 20-05-30. Download link: sid-00603-sp-k2048-nms3-refine2-r1600forcecubic-down128-masked-d.001_sg-t.2-it150_degensac-th1.2.json
This page ranks the submission against all others using the same number of keypoints, regardless of descriptor size. Please hover over table headers for descriptions on metrics and full scene names.
Metadata
- Authors: Paul-Edouard Sarlin (contact)
- Keypoint: superpoint-k2048-nms3-refine2-r1600forcecubic-masked-d.001
- Descriptor: superpoint-down128 (128 float32: 512 bytes)
- Number of features: 2048
- Summary: SuperPoint detector (2048 keypoints, NMS with radius 3, confidence threshold 0.001, refinement, on 1600-pixel images). Detections on semantic classes sky and people are removed (segmentation from HFNetV2 trained on MIT ADE20K). SuperPoint descriptor, reduced to 128d with a linear autoencoder. SuperGlue matcher (outdoor model, 150 Sinkhorn iterations). For stereo, DEGENSAC model estimator (1.2 pixel inlier threshold).
- Paper: https://arxiv.org/abs/1911.11763
- Website: https://psarlin.com/superglue
- Origin: Submission
- Flags: is_submission, is_challenge_2020
Phototourism / Stereo track
mAA at 10 degrees: 0.56769 (±0.00034 over 3 run(s) / ±0.13842 over 9 scenes)
Rank (per category): 3 (of 108)
Scene | Features | Matches (matcher) |
Matches (filter) |
Matches (final) |
Rep. @ 3 px. | MS @ 3 px. | mAA(5o) | mAA(10o) |
bm | 2035.9 | — | — | 334.3 | 0.425 Rank: 82/108 |
0.877 Rank: 71/108 |
0.33883 (±0.00311) Rank: 3/108 |
0.49414 (±0.00181) Rank: 3/108 |
fcs | 2048.0 | — | — | 502.1 | 0.407 Rank: 17/108 |
0.860 Rank: 60/108 |
0.64122 (±0.00042) Rank: 3/108 |
0.76250 (±0.00065) Rank: 3/108 |
lms | 2048.0 | — | — | 422.9 | 0.380 Rank: 40/108 |
0.674 Rank: 66/108 |
0.59865 (±0.00132) Rank: 4/108 |
0.72362 (±0.00091) Rank: 4/108 |
lb | 2045.2 | — | — | 380.2 | 0.386 Rank: 21/108 |
0.689 Rank: 43/108 |
0.49033 (±0.00311) Rank: 3/108 |
0.60681 (±0.00070) Rank: 3/108 |
mc | 2048.0 | — | — | 389.2 | 0.402 Rank: 28/108 |
0.841 Rank: 76/108 |
0.32015 (±0.00311) Rank: 3/108 |
0.47291 (±0.00148) Rank: 16/108 |
mr | 2000.6 | — | — | 426.6 | 0.452 Rank: 16/108 |
0.888 Rank: 69/108 |
0.25087 (±0.00152) Rank: 10/108 |
0.36269 (±0.00090) Rank: 7/108 |
psm | 2048.0 | — | — | 325.6 | 0.305 Rank: 31/108 |
0.581 Rank: 10/108 |
0.22452 (±0.00263) Rank: 3/108 |
0.37607 (±0.00353) Rank: 3/108 |
sf | 2048.0 | — | — | 475.6 | 0.365 Rank: 24/108 |
0.773 Rank: 73/108 |
0.50117 (±0.00242) Rank: 3/108 |
0.64568 (±0.00146) Rank: 3/108 |
spc | 2048.0 | — | — | 385.9 | 0.364 Rank: 28/108 |
0.781 Rank: 67/108 |
0.49747 (±0.00308) Rank: 3/108 |
0.66484 (±0.00164) Rank: 3/108 |
avg | 2041.1 | — | — | 404.7 | 0.387 Rank: 25/108 |
0.774 Rank: 55/108 |
0.42925 (±0.00030) Rank: 3/108 |
0.56769 (±0.00034) Rank: 3/108 |
We show the inliers that survive the robust estimation loop (i.e. RANSAC), or those supplied with the submission if using custom matches, and use the depth estimates to determine whether they are correct. We draw matches above a 5-pixel error threshold in red, and those below are color-coded by their error, from 0 (green) to 5 pixels (yellow). Matches for which we do not have depth estimates are drawn in blue. Please note that the depth maps are estimates and may contain errors.
— british museum —
— florence cathedral side —
— lincoln memorial statue —
— london bridge —
— milan cathedral —
— mount rushmore —
— piazza san marco —
— sagrada familia —
— saint paul's cathedral —
Phototourism / Multiview track
mAA at 10 degrees: 0.76987 (±0.00122 over 3 run(s) / ±0.10723 over 9 scenes)
Rank (per category): 3 (of 108)
Scene | Features | Matches (input) |
RegistrationRatio (%) | Number of Landmarks |
Track Length | ATE | mAA(50) | mAA(100) |
bm | 2035.9 | 334.29 | 99.91 Rank: 12/108 |
1914.52 Rank: 21/108 |
4.622 Rank: 76/108 |
0.36096 Rank: 4/108 |
0.57935 (±0.00220) Rank: 5/108 |
0.72044 (±0.00331) Rank: 6/108 |
fcs | 2048.0 | 508.23 | 97.98 Rank: 19/108 |
2561.68 Rank: 7/108 |
5.038 Rank: 10/108 |
0.23423 Rank: 8/108 |
0.75273 (±0.00549) Rank: 3/108 |
0.81000 (±0.00514) Rank: 6/108 |
lms | 2048.0 | 442.56 | 99.49 Rank: 5/108 |
1963.42 Rank: 24/108 |
5.123 Rank: 22/108 |
0.24860 Rank: 1/108 |
0.86579 (±0.00186) Rank: 2/108 |
0.91480 (±0.00122) Rank: 2/108 |
lb | 2045.2 | 464.82 | 98.79 Rank: 1/108 |
2004.77 Rank: 15/108 |
5.320 Rank: 12/108 |
0.49328 Rank: 27/108 |
0.70707 (±0.00131) Rank: 10/108 |
0.80629 (±0.00119) Rank: 11/108 |
mc | 2048.0 | 384.11 | 99.88 Rank: 15/108 |
2054.16 Rank: 16/108 |
5.009 Rank: 15/108 |
0.34594 Rank: 9/108 |
0.56161 (±0.00064) Rank: 3/108 |
0.70723 (±0.00044) Rank: 2/108 |
mr | 2000.6 | 418.86 | 94.08 Rank: 16/108 |
1785.44 Rank: 28/108 |
4.922 Rank: 4/108 |
0.48607 Rank: 2/108 |
0.40395 (±0.00746) Rank: 6/108 |
0.52885 (±0.00963) Rank: 5/108 |
psm | 2048.0 | 324.07 | 98.56 Rank: 10/108 |
2325.25 Rank: 17/108 |
4.355 Rank: 7/108 |
0.32425 Rank: 4/108 |
0.65242 (±0.00636) Rank: 9/108 |
0.73579 (±0.00708) Rank: 8/108 |
sf | 2048.0 | 468.55 | 99.82 Rank: 10/108 |
2645.18 Rank: 8/108 |
4.974 Rank: 9/108 |
0.28238 Rank: 4/108 |
0.78041 (±0.00203) Rank: 3/108 |
0.86256 (±0.00177) Rank: 3/108 |
spc | 2048.0 | 396.25 | 100.00 Rank: 1/108 |
2167.87 Rank: 16/108 |
5.022 Rank: 11/108 |
0.43712 Rank: 12/108 |
0.74555 (±0.00173) Rank: 6/108 |
0.84288 (±0.00155) Rank: 4/108 |
avg | 2041.1 | 415.75 | 98.72 Rank: 8/108 |
2158.03 Rank: 15/108 |
4.932 Rank: 12/108 |
0.35698 Rank: 1/108 |
0.67210 (±0.00096) Rank: 3/108 |
0.76987 (±0.00122) Rank: 3/108 |
In the multi-view track we reconstruct the scene with Structure-from-Motion (Colmap) with small sets of images. We show the results for one bag of 25 images (displaying: 10). Keypoints are drawn in blue if they are part of the model, and in red otherwise.
— british museum —
— florence cathedral side —
— lincoln memorial statue —
— london bridge —
— milan cathedral —
— mount rushmore —
— piazza san marco —
— sagrada familia —
— saint paul's cathedral —