Submission ID: 00646
MT-2-Hardnet-Pretraind-all-Datasets-512-bytes
Processed: 21-03-28. Download link: sid-00646-mt-2-hardnet-pretrained-all-datasets-magsac.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: Anonymous (to be released: 2020-6-12)
- Keypoint: mt-2
- Descriptor: hardnet (128 float32: 512 bytes)
- Number of features: 2048
- Summary: Local feature model learned via training with covariant constraint loss function. We take the top 2048 features score-wise. HardNet, pre-trained on all datasets, is used as a descriptor head. Graph-Cut(GC)-RANSAC is used as a robust estimator. Cyclic consistency matching with a threshold of 0.95 is used.
- Paper: Anonymous (to be released: 2020-6-12)
- Website: Anonymous (to be released: 2020-6-12)
- Origin: Submission
- Flags: is_submission
Phototourism / Stereo track
mAA at 10 degrees: 0.32109 (±0.00043 over 3 run(s) / ±0.15566 over 9 scenes)
Rank (per category): 85 (of 108)
| Scene | Features | Matches (matcher) |
Matches (filter) |
Matches (final) |
Rep. @ 3 px. | MS @ 3 px. | mAA(5o) | mAA(10o) |
| bm | 2048.0 | 310.1 | 310.1 | 262.6 | 0.592 Rank: 3/108 |
0.713 Rank: 96/108 |
0.02714 (±0.00038) Rank: 93/108 |
0.05925 (±0.00086) Rank: 93/108 |
| fcs | 2048.0 | 261.2 | 261.2 | 211.5 | 0.374 Rank: 29/108 |
0.808 Rank: 81/108 |
0.39736 (±0.00082) Rank: 84/108 |
0.54690 (±0.00077) Rank: 83/108 |
| lms | 2048.0 | 340.2 | 340.2 | 289.3 | 0.381 Rank: 36/108 |
0.582 Rank: 90/108 |
0.40978 (±0.00162) Rank: 48/108 |
0.54478 (±0.00141) Rank: 42/108 |
| lb | 2048.0 | 265.9 | 265.9 | 197.2 | 0.409 Rank: 10/108 |
0.587 Rank: 93/108 |
0.16732 (±0.00038) Rank: 93/108 |
0.23613 (±0.00402) Rank: 92/108 |
| mc | 2048.0 | 286.0 | 286.0 | 229.7 | 0.509 Rank: 16/108 |
0.839 Rank: 78/108 |
0.19018 (±0.00038) Rank: 93/108 |
0.31031 (±0.00267) Rank: 78/108 |
| mr | 2048.0 | 373.6 | 373.6 | 334.5 | 0.479 Rank: 5/108 |
0.847 Rank: 81/108 |
0.17509 (±0.00095) Rank: 69/108 |
0.25611 (±0.00069) Rank: 75/108 |
| psm | 2048.0 | 153.2 | 153.2 | 121.0 | 0.379 Rank: 3/108 |
0.390 Rank: 66/108 |
0.07655 (±0.00125) Rank: 76/108 |
0.15993 (±0.00165) Rank: 76/108 |
| sf | 2048.0 | 376.1 | 376.1 | 326.0 | 0.449 Rank: 2/108 |
0.741 Rank: 81/108 |
0.28199 (±0.00061) Rank: 80/108 |
0.41927 (±0.00092) Rank: 79/108 |
| spc | 2048.0 | 263.3 | 263.3 | 204.8 | 0.479 Rank: 1/108 |
0.695 Rank: 83/108 |
0.21686 (±0.00151) Rank: 84/108 |
0.35715 (±0.00081) Rank: 83/108 |
| avg | 2048.0 | 292.2 | 292.2 | 241.8 | 0.450 Rank: 4/108 |
0.689 Rank: 86/108 |
0.21581 (±0.00022) Rank: 84/108 |
0.32109 (±0.00043) Rank: 85/108 |
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: 0.60774 (±0.00266 over 3 run(s) / ±0.18722 over 9 scenes)
Rank (per category): 54 (of 108)
| Scene | Features | Matches (input) |
RegistrationRatio (%) | Number of Landmarks |
Track Length | ATE | mAA(50) | mAA(100) |
| bm | 2048.0 | 308.98 | 99.60 Rank: 59/108 |
1474.75 Rank: 40/108 |
4.511 Rank: 84/108 |
0.86235 Rank: 89/108 |
0.15133 (±0.00301) Rank: 92/108 |
0.25603 (±0.00449) Rank: 92/108 |
| fcs | 2048.0 | 269.42 | 98.42 Rank: 7/108 |
1724.66 Rank: 58/108 |
4.047 Rank: 77/108 |
0.25891 Rank: 29/108 |
0.70969 (±0.00938) Rank: 29/108 |
0.78764 (±0.00946) Rank: 22/108 |
| lms | 2048.0 | 368.35 | 98.84 Rank: 38/108 |
1888.82 Rank: 29/108 |
4.460 Rank: 85/108 |
0.32189 Rank: 59/108 |
0.73919 (±0.00094) Rank: 64/108 |
0.82054 (±0.00404) Rank: 58/108 |
| lb | 2048.0 | 323.44 | 98.35 Rank: 29/108 |
1429.46 Rank: 39/108 |
4.514 Rank: 43/108 |
0.52055 Rank: 59/108 |
0.49005 (±0.01304) Rank: 60/108 |
0.60432 (±0.01344) Rank: 68/108 |
| mc | 2048.0 | 280.36 | 99.66 Rank: 28/108 |
1280.31 Rank: 48/108 |
4.576 Rank: 35/108 |
0.42724 Rank: 36/108 |
0.45457 (±0.00557) Rank: 38/108 |
0.60934 (±0.00775) Rank: 36/108 |
| mr | 2048.0 | 373.89 | 95.12 Rank: 4/108 |
1675.35 Rank: 42/108 |
4.492 Rank: 25/108 |
0.53473 Rank: 26/108 |
0.36483 (±0.00971) Rank: 28/108 |
0.49224 (±0.01261) Rank: 22/108 |
| psm | 2048.0 | 148.70 | 87.10 Rank: 37/108 |
1451.37 Rank: 47/108 |
3.037 Rank: 55/108 |
0.65906 Rank: 39/108 |
0.28824 (±0.00466) Rank: 41/108 |
0.37769 (±0.00911) Rank: 38/108 |
| sf | 2048.0 | 371.50 | 98.16 Rank: 26/108 |
1853.29 Rank: 49/108 |
4.561 Rank: 35/108 |
0.33400 Rank: 44/108 |
0.68639 (±0.00185) Rank: 36/108 |
0.78235 (±0.00194) Rank: 33/108 |
| spc | 2048.0 | 270.15 | 99.49 Rank: 31/108 |
1559.74 Rank: 48/108 |
4.363 Rank: 44/108 |
0.55635 Rank: 66/108 |
0.61855 (±0.00535) Rank: 62/108 |
0.73951 (±0.00559) Rank: 54/108 |
| avg | 2048.0 | 301.64 | 97.19 Rank: 31/108 |
1593.08 Rank: 44/108 |
4.285 Rank: 59/108 |
0.49723 Rank: 59/108 |
0.50032 (±0.00185) Rank: 56/108 |
0.60774 (±0.00266) Rank: 54/108 |
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 —
