Submission ID: 00711
DISK-32D (depth)
Processed: 20-09-16. Download link: sid-00711-disk-2k-32D-depth.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: Michal Tyszkiewicz (contact)
- Keypoint: disk-2020-09-15-nms-7-depth-32-save-46-imsize-1024-nump-2048
- Descriptor: disk-2020-09-15-nms-7-depth-32-save-46-imsize-1024-nump-2048 (32 float32: 128 bytes)
- Number of features: 2048
- Summary: Local feature model learned via policy gradient, using 32D descriptors. Model trained with a cycle-consistency loss and supervised with depth-based constraints. Trained on MegaDepth, removing conflicts with the test data. For inference, images are resized to 1024 pixels on the longest edge, with NMS over a 7x7 window, taking the top 2048 features by score.
- Paper: N/A
- Website: N/A
- Origin: Baseline
- Flags: is_baseline
Phototourism / Stereo track
mAA at 10 degrees: 0.47005 (±0.00037 over 3 run(s) / ±0.11981 over 9 scenes)
Rank (per category): 15 (of 108)
| Scene | Features | Matches (matcher) |
Matches (filter) |
Matches (final) |
Rep. @ 3 px. | MS @ 3 px. | mAA(5o) | mAA(10o) |
| bm | 2048.0 | 695.0 | 695.0 | 552.2 | 0.601 Rank: 2/108 |
0.958 Rank: 2/108 |
0.24510 (±0.00057) Rank: 16/108 |
0.38091 (±0.00181) Rank: 16/108 |
| fcs | 2048.0 | 428.6 | 428.6 | 340.0 | 0.409 Rank: 16/108 |
0.900 Rank: 17/108 |
0.53492 (±0.00225) Rank: 23/108 |
0.65912 (±0.00200) Rank: 28/108 |
| lms | 2048.0 | 269.5 | 269.5 | 215.5 | 0.356 Rank: 85/108 |
0.716 Rank: 2/108 |
0.38503 (±0.00192) Rank: 64/108 |
0.48947 (±0.00084) Rank: 69/108 |
| lb | 2048.0 | 404.6 | 404.6 | 320.9 | 0.496 Rank: 2/108 |
0.780 Rank: 2/108 |
0.44523 (±0.00057) Rank: 16/108 |
0.54820 (±0.00129) Rank: 14/108 |
| mc | 2048.0 | 580.1 | 580.1 | 478.8 | 0.530 Rank: 3/108 |
0.964 Rank: 1/108 |
0.34111 (±0.00057) Rank: 16/108 |
0.49986 (±0.00154) Rank: 8/108 |
| mr | 2048.0 | 375.4 | 375.4 | 314.5 | 0.414 Rank: 27/108 |
0.905 Rank: 55/108 |
0.25374 (±0.00071) Rank: 8/108 |
0.34666 (±0.00076) Rank: 17/108 |
| psm | 2048.0 | 254.9 | 254.9 | 197.4 | 0.364 Rank: 13/108 |
0.684 Rank: 2/108 |
0.12648 (±0.00063) Rank: 25/108 |
0.23872 (±0.00103) Rank: 25/108 |
| sf | 2048.0 | 365.0 | 365.0 | 286.7 | 0.392 Rank: 14/108 |
0.810 Rank: 43/108 |
0.38766 (±0.00047) Rank: 28/108 |
0.50863 (±0.00069) Rank: 37/108 |
| spc | 2048.0 | 481.6 | 481.6 | 370.4 | 0.473 Rank: 7/108 |
0.865 Rank: 11/108 |
0.40083 (±0.00026) Rank: 19/108 |
0.55889 (±0.00102) Rank: 19/108 |
| avg | 2048.0 | 428.3 | 428.3 | 341.8 | 0.448 Rank: 10/108 |
0.842 Rank: 3/108 |
0.34668 (±0.00022) Rank: 15/108 |
0.47005 (±0.00037) Rank: 15/108 |
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: 0.70220 (±0.00208 over 3 run(s) / ±0.10063 over 9 scenes)
Rank (per category): 19 (of 108)
| Scene | Features | Matches (input) |
RegistrationRatio (%) | Number of Landmarks |
Track Length | ATE | mAA(50) | mAA(100) |
| bm | 2048.0 | 698.85 | 99.98 Rank: 6/108 |
2293.14 Rank: 5/108 |
6.747 Rank: 2/108 |
0.41073 Rank: 14/108 |
0.52197 (±0.00379) Rank: 17/108 |
0.67215 (±0.00400) Rank: 16/108 |
| fcs | 2048.0 | 442.40 | 96.49 Rank: 58/108 |
2427.55 Rank: 13/108 |
5.203 Rank: 6/108 |
0.29720 Rank: 77/108 |
0.67082 (±0.01004) Rank: 53/108 |
0.74188 (±0.01133) Rank: 64/108 |
| lms | 2048.0 | 292.39 | 98.74 Rank: 41/108 |
1496.11 Rank: 48/108 |
5.510 Rank: 10/108 |
0.37019 Rank: 87/108 |
0.73888 (±0.01004) Rank: 65/108 |
0.81131 (±0.01246) Rank: 71/108 |
| lb | 2048.0 | 490.55 | 98.59 Rank: 21/108 |
1972.37 Rank: 18/108 |
6.374 Rank: 2/108 |
0.47548 Rank: 6/108 |
0.69543 (±0.00224) Rank: 12/108 |
0.79054 (±0.00186) Rank: 16/108 |
| mc | 2048.0 | 569.50 | 99.78 Rank: 25/108 |
2397.44 Rank: 9/108 |
6.128 Rank: 5/108 |
0.38321 Rank: 20/108 |
0.52802 (±0.00148) Rank: 17/108 |
0.67226 (±0.00221) Rank: 17/108 |
| mr | 2048.0 | 368.89 | 92.71 Rank: 38/108 |
2336.73 Rank: 3/108 |
4.788 Rank: 12/108 |
0.51493 Rank: 8/108 |
0.39356 (±0.00967) Rank: 11/108 |
0.50486 (±0.01300) Rank: 14/108 |
| psm | 2048.0 | 252.62 | 93.06 Rank: 22/108 |
2359.56 Rank: 16/108 |
4.212 Rank: 13/108 |
0.53899 Rank: 23/108 |
0.48713 (±0.01243) Rank: 20/108 |
0.56999 (±0.01177) Rank: 21/108 |
| sf | 2048.0 | 360.83 | 95.34 Rank: 70/108 |
2425.78 Rank: 14/108 |
4.966 Rank: 10/108 |
0.32367 Rank: 33/108 |
0.70104 (±0.00459) Rank: 28/108 |
0.77808 (±0.00456) Rank: 37/108 |
| spc | 2048.0 | 494.45 | 98.53 Rank: 57/108 |
2375.26 Rank: 8/108 |
5.792 Rank: 6/108 |
0.42365 Rank: 7/108 |
0.67301 (±0.00235) Rank: 33/108 |
0.77878 (±0.00225) Rank: 36/108 |
| avg | 2048.0 | 441.16 | 97.02 Rank: 34/108 |
2231.55 Rank: 11/108 |
5.524 Rank: 6/108 |
0.41534 Rank: 16/108 |
0.60109 (±0.00187) Rank: 16/108 |
0.70220 (±0.00208) Rank: 19/108 |
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 —
