Submission ID: 00599
SEKD
Processed: 20-05-29. Download link: sid-00599-sekd-2k-both-degensac.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: Yafei Song, Ling Cai, Jia Li, Yonghong Tian, Mingyang Li (contact)
- Keypoint: sekd
- Descriptor: sekd (128 float32: 512 bytes)
- Number of features: 2048
- Summary: Our method named SEKD: Self-Evolving Keypoint Detection and Description, where the SEKD model is trained using COCO validation set. In this submission each image has up to 2048 SEKD keypoints, and 128-dim float descriptor. We use the built-in matcher (bidirectional filter with the both strategy, without flann, optimal inlier and ratio test thresholds) with DEGENSAC.
- Paper: N/A
- Website: N/A
- Origin: Submission
- Flags: is_submission, is_challenge_2020
Phototourism / Stereo track
mAA at 10 degrees: 0.45066 (±0.00057 over 3 run(s) / ±0.14265 over 9 scenes)
Rank (per category): 26 (of 108)
| Scene | Features | Matches (matcher) |
Matches (filter) |
Matches (final) |
Rep. @ 3 px. | MS @ 3 px. | mAA(5o) | mAA(10o) |
| bm | 2048.0 | 225.8 | 225.8 | 125.4 | 0.495 Rank: 22/108 |
0.930 Rank: 40/108 |
0.22851 (±0.00017) Rank: 20/108 |
0.35202 (±0.00082) Rank: 20/108 |
| fcs | 2043.4 | 237.7 | 237.7 | 140.9 | 0.369 Rank: 35/108 |
0.903 Rank: 13/108 |
0.54388 (±0.00400) Rank: 20/108 |
0.67574 (±0.00307) Rank: 18/108 |
| lms | 2045.1 | 329.9 | 329.9 | 203.3 | 0.396 Rank: 15/108 |
0.699 Rank: 34/108 |
0.49638 (±0.00050) Rank: 18/108 |
0.62150 (±0.00038) Rank: 17/108 |
| lb | 2047.2 | 202.8 | 202.8 | 103.7 | 0.366 Rank: 25/108 |
0.735 Rank: 9/108 |
0.34334 (±0.00017) Rank: 20/108 |
0.45102 (±0.00075) Rank: 26/108 |
| mc | 2043.6 | 210.5 | 210.5 | 120.1 | 0.413 Rank: 25/108 |
0.910 Rank: 40/108 |
0.25908 (±0.00017) Rank: 20/108 |
0.40129 (±0.00123) Rank: 49/108 |
| mr | 2045.2 | 242.3 | 242.3 | 143.6 | 0.368 Rank: 40/108 |
0.913 Rank: 48/108 |
0.23502 (±0.00082) Rank: 24/108 |
0.33118 (±0.00174) Rank: 24/108 |
| psm | 2037.9 | 116.8 | 116.8 | 60.6 | 0.332 Rank: 24/108 |
0.519 Rank: 26/108 |
0.10554 (±0.00148) Rank: 36/108 |
0.19681 (±0.00160) Rank: 38/108 |
| sf | 2046.1 | 279.8 | 279.8 | 165.8 | 0.351 Rank: 36/108 |
0.848 Rank: 8/108 |
0.41714 (±0.00262) Rank: 15/108 |
0.56105 (±0.00212) Rank: 14/108 |
| spc | 2034.6 | 200.6 | 200.6 | 101.9 | 0.383 Rank: 24/108 |
0.820 Rank: 45/108 |
0.31871 (±0.00309) Rank: 56/108 |
0.46537 (±0.00271) Rank: 55/108 |
| avg | 2043.4 | 227.4 | 227.4 | 129.5 | 0.386 Rank: 26/108 |
0.809 Rank: 20/108 |
0.32751 (±0.00102) Rank: 25/108 |
0.45066 (±0.00057) Rank: 26/108 |
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: 0.66094 (±0.00227 over 3 run(s) / ±0.15512 over 9 scenes)
Rank (per category): 33 (of 108)
| Scene | Features | Matches (input) |
RegistrationRatio (%) | Number of Landmarks |
Track Length | ATE | mAA(50) | mAA(100) |
| bm | 2048.0 | 169.29 | 99.93 Rank: 9/108 |
841.97 Rank: 98/108 |
4.903 Rank: 59/108 |
0.48568 Rank: 20/108 |
0.45420 (±0.01233) Rank: 20/108 |
0.60411 (±0.01332) Rank: 20/108 |
| fcs | 2043.4 | 185.85 | 97.16 Rank: 48/108 |
1457.90 Rank: 87/108 |
4.323 Rank: 42/108 |
0.28814 Rank: 68/108 |
0.69946 (±0.01086) Rank: 36/108 |
0.77276 (±0.01031) Rank: 37/108 |
| lms | 2045.1 | 285.46 | 99.56 Rank: 2/108 |
1350.87 Rank: 56/108 |
5.323 Rank: 15/108 |
0.29750 Rank: 25/108 |
0.82455 (±0.00489) Rank: 10/108 |
0.88570 (±0.00469) Rank: 12/108 |
| lb | 2047.2 | 178.62 | 97.97 Rank: 41/108 |
1015.08 Rank: 90/108 |
4.800 Rank: 31/108 |
0.49120 Rank: 23/108 |
0.62992 (±0.00977) Rank: 26/108 |
0.73831 (±0.01050) Rank: 28/108 |
| mc | 2043.6 | 150.10 | 98.30 Rank: 41/108 |
1045.81 Rank: 78/108 |
4.619 Rank: 32/108 |
0.44548 Rank: 44/108 |
0.44924 (±0.00662) Rank: 41/108 |
0.59387 (±0.00978) Rank: 40/108 |
| mr | 2045.2 | 190.19 | 91.46 Rank: 62/108 |
1362.86 Rank: 76/108 |
4.180 Rank: 42/108 |
0.53248 Rank: 23/108 |
0.36597 (±0.00534) Rank: 25/108 |
0.48053 (±0.00717) Rank: 31/108 |
| psm | 2037.9 | 69.37 | 81.58 Rank: 53/108 |
861.80 Rank: 93/108 |
3.127 Rank: 39/108 |
0.67110 Rank: 41/108 |
0.30233 (±0.01325) Rank: 37/108 |
0.36496 (±0.01160) Rank: 40/108 |
| sf | 2046.1 | 215.08 | 97.24 Rank: 34/108 |
1751.47 Rank: 68/108 |
4.382 Rank: 65/108 |
0.33290 Rank: 42/108 |
0.70532 (±0.00472) Rank: 23/108 |
0.78809 (±0.00583) Rank: 28/108 |
| spc | 2034.6 | 145.46 | 97.72 Rank: 69/108 |
1198.89 Rank: 88/108 |
4.272 Rank: 63/108 |
0.53802 Rank: 52/108 |
0.61904 (±0.00112) Rank: 59/108 |
0.72018 (±0.00179) Rank: 66/108 |
| avg | 2043.4 | 176.60 | 95.66 Rank: 49/108 |
1209.63 Rank: 87/108 |
4.437 Rank: 36/108 |
0.45361 Rank: 32/108 |
0.56111 (±0.00177) Rank: 31/108 |
0.66094 (±0.00227) Rank: 33/108 |
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 —
