Submission ID: 00527
SEKD
Processed: 20-04-27. Download link: sid-00527-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, Mingyang Li (contact)
- Keypoint: sekd
- Descriptor: sekd (128 float32: 512 bytes)
- Number of features: 2048
- Summary: The name of our method is Self-Evolving Keypoint Detection and Description (SEKD). Now, the SEKD model is trained only using COCO test images. 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, 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: 0.38886 (±0.00013 over 3 run(s) / ±0.13090 over 9 scenes)
Rank (per category): 59 (of 108)
| Scene | Features | Matches (matcher) |
Matches (filter) |
Matches (final) |
Rep. @ 3 px. | MS @ 3 px. | mAA(5o) | mAA(10o) |
| bm | 1688.9 | 182.6 | 182.6 | 102.9 | 0.494 Rank: 24/108 |
0.906 Rank: 65/108 |
0.18479 (±0.00193) Rank: 34/108 |
0.29326 (±0.00155) Rank: 39/108 |
| fcs | 1969.0 | 218.6 | 218.6 | 126.9 | 0.368 Rank: 37/108 |
0.867 Rank: 57/108 |
0.49246 (±0.00324) Rank: 55/108 |
0.63123 (±0.00302) Rank: 43/108 |
| lms | 845.4 | 198.3 | 198.3 | 137.2 | 0.436 Rank: 6/108 |
0.663 Rank: 74/108 |
0.37935 (±0.00159) Rank: 69/108 |
0.50172 (±0.00226) Rank: 63/108 |
| lb | 1806.0 | 163.0 | 163.0 | 84.1 | 0.351 Rank: 31/108 |
0.689 Rank: 42/108 |
0.27201 (±0.00193) Rank: 34/108 |
0.37555 (±0.00148) Rank: 59/108 |
| mc | 1964.4 | 186.6 | 186.6 | 104.2 | 0.404 Rank: 27/108 |
0.848 Rank: 75/108 |
0.21356 (±0.00193) Rank: 34/108 |
0.34435 (±0.00235) Rank: 69/108 |
| mr | 1938.7 | 192.0 | 192.0 | 113.5 | 0.349 Rank: 45/108 |
0.816 Rank: 83/108 |
0.18467 (±0.00211) Rank: 62/108 |
0.26889 (±0.00312) Rank: 66/108 |
| psm | 1963.3 | 107.5 | 107.5 | 56.3 | 0.326 Rank: 27/108 |
0.478 Rank: 37/108 |
0.09192 (±0.00129) Rank: 53/108 |
0.17756 (±0.00074) Rank: 54/108 |
| sf | 1998.8 | 247.1 | 247.1 | 143.9 | 0.342 Rank: 41/108 |
0.797 Rank: 55/108 |
0.35434 (±0.00302) Rank: 53/108 |
0.49837 (±0.00357) Rank: 48/108 |
| spc | 1905.7 | 173.6 | 173.6 | 88.1 | 0.375 Rank: 27/108 |
0.764 Rank: 76/108 |
0.26694 (±0.00162) Rank: 78/108 |
0.40886 (±0.00024) Rank: 75/108 |
| avg | 1786.7 | 185.5 | 185.5 | 106.3 | 0.383 Rank: 29/108 |
0.759 Rank: 71/108 |
0.27111 (±0.00050) Rank: 65/108 |
0.38886 (±0.00013) Rank: 59/108 |
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: 0.63303 (±0.00244 over 3 run(s) / ±0.14792 over 9 scenes)
Rank (per category): 45 (of 108)
| Scene | Features | Matches (input) |
RegistrationRatio (%) | Number of Landmarks |
Track Length | ATE | mAA(50) | mAA(100) |
| bm | 1688.9 | 144.02 | 99.47 Rank: 76/108 |
715.51 Rank: 104/108 |
4.994 Rank: 53/108 |
0.51409 Rank: 22/108 |
0.41950 (±0.00398) Rank: 23/108 |
0.56635 (±0.00532) Rank: 23/108 |
| fcs | 1969.0 | 177.49 | 97.10 Rank: 52/108 |
1405.48 Rank: 93/108 |
4.400 Rank: 32/108 |
0.31537 Rank: 86/108 |
0.70030 (±0.00765) Rank: 35/108 |
0.77322 (±0.00571) Rank: 36/108 |
| lms | 845.4 | 182.46 | 98.66 Rank: 42/108 |
694.64 Rank: 105/108 |
6.225 Rank: 2/108 |
0.32651 Rank: 66/108 |
0.74051 (±0.00444) Rank: 60/108 |
0.81624 (±0.00341) Rank: 66/108 |
| lb | 1806.0 | 151.94 | 97.65 Rank: 47/108 |
888.62 Rank: 99/108 |
4.859 Rank: 30/108 |
0.49823 Rank: 36/108 |
0.58617 (±0.00588) Rank: 35/108 |
0.70017 (±0.00517) Rank: 37/108 |
| mc | 1964.4 | 140.04 | 97.39 Rank: 53/108 |
1005.38 Rank: 86/108 |
4.667 Rank: 30/108 |
0.47834 Rank: 59/108 |
0.43023 (±0.00681) Rank: 51/108 |
0.57117 (±0.00870) Rank: 49/108 |
| mr | 1938.7 | 156.36 | 90.31 Rank: 71/108 |
1262.17 Rank: 86/108 |
4.114 Rank: 44/108 |
0.53810 Rank: 30/108 |
0.34712 (±0.00485) Rank: 46/108 |
0.45580 (±0.00576) Rank: 53/108 |
| psm | 1963.3 | 67.16 | 79.16 Rank: 62/108 |
835.33 Rank: 95/108 |
3.128 Rank: 37/108 |
0.64645 Rank: 36/108 |
0.29719 (±0.00678) Rank: 40/108 |
0.35477 (±0.00660) Rank: 44/108 |
| sf | 1998.8 | 199.63 | 96.06 Rank: 57/108 |
1649.95 Rank: 84/108 |
4.451 Rank: 52/108 |
0.32217 Rank: 29/108 |
0.68758 (±0.00123) Rank: 35/108 |
0.77079 (±0.00254) Rank: 41/108 |
| spc | 1905.7 | 131.85 | 97.31 Rank: 75/108 |
1110.64 Rank: 99/108 |
4.335 Rank: 52/108 |
0.54949 Rank: 60/108 |
0.58509 (±0.00641) Rank: 70/108 |
0.68874 (±0.00639) Rank: 72/108 |
| avg | 1786.7 | 150.11 | 94.79 Rank: 61/108 |
1063.08 Rank: 100/108 |
4.575 Rank: 23/108 |
0.46542 Rank: 42/108 |
0.53263 (±0.00269) Rank: 40/108 |
0.63303 (±0.00244) Rank: 45/108 |
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 —
