Submission ID: 7b1769c7
KORNIA TUTORIAL CV-DoG-AffNet-HardNet8
Processed: 21-05-15. Download link: 7b1769c7167d5bad-dog-affnet-hardnet8-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.
- Phototourism dataset: Stereo track / Multiview track
- Prague Parks dataset: Stereo track / Multiview track
- Google Urban dataset: Stereo track / Multiview track
Metadata
- Authors: Dmytro Mishkin, Milan Pultar and kornia team (contact)
- Keypoint: cv2dog
- Descriptor: affnethardnet8 (128 float32: 512 bytes)
- Number of features: 8000
- Summary: OpeCV SIFT keypoints 8000 features, followed by the AffNet normalization and HardNet8 descriptor as implemented in kornia. Matched using the built-in matcher (bidirectional filter with the both strategy, hopefully optimal inlier and ratio test thresholds) with DEGENSAC
- Paper: https://arxiv.org/abs/2007.09699
- Website: https://github.com/kornia/kornia
- Processing date: 21-05-15
Phototourism dataset / Stereo track
mAA at 10 degrees: 0.55738 (±0.00000 over 1 run(s) / ±0.14399 over 9 scenes)
Rank (per category): 19 (of 51)
| Scene | Features | Matches (raw) |
Matches (final) |
Rep. @ 3 px. | MS @ 3 px. | mAA(5o) | mAA(10o) |
| BM | 7969.3 | 596.1 | 405.8 | 0.526 Rank: 38/51 |
0.946 Rank: 13/51 |
0.22733 (±0.00000) Rank: 30/51 |
0.34565 (±0.00000) Rank: 30/51 |
| FCS | 7876.6 | 460.3 | 323.7 | 0.443 Rank: 39/51 |
0.944 Rank: 2/51 |
0.63138 (±0.00000) Rank: 24/51 |
0.74707 (±0.00000) Rank: 25/51 |
| LMS | 7773.0 | 381.9 | 200.5 | 0.465 Rank: 28/51 |
0.681 Rank: 25/51 |
0.54552 (±0.00000) Rank: 25/51 |
0.67808 (±0.00000) Rank: 24/51 |
| LB | 7958.1 | 352.6 | 217.8 | 0.472 Rank: 37/51 |
0.805 Rank: 2/51 |
0.52184 (±0.00000) Rank: 30/51 |
0.62384 (±0.00000) Rank: 17/51 |
| MC | 7828.2 | 538.9 | 373.6 | 0.525 Rank: 43/51 |
0.965 Rank: 4/51 |
0.40365 (±0.00000) Rank: 30/51 |
0.55518 (±0.00000) Rank: 21/51 |
| MR | 7795.9 | 533.6 | 388.0 | 0.457 Rank: 43/51 |
0.977 Rank: 2/51 |
0.33534 (±0.00000) Rank: 13/51 |
0.44787 (±0.00000) Rank: 13/51 |
| PSM | 7811.8 | 259.7 | 167.8 | 0.385 Rank: 46/51 |
0.778 Rank: 6/51 |
0.18570 (±0.00000) Rank: 22/51 |
0.31771 (±0.00000) Rank: 19/51 |
| SF | 7900.7 | 675.1 | 458.0 | 0.479 Rank: 36/51 |
0.898 Rank: 2/51 |
0.52012 (±0.00000) Rank: 25/51 |
0.65342 (±0.00000) Rank: 23/51 |
| SPC | 7871.1 | 540.6 | 349.2 | 0.487 Rank: 40/51 |
0.903 Rank: 1/51 |
0.49252 (±0.00000) Rank: 16/51 |
0.64761 (±0.00000) Rank: 15/51 |
| Avg | 7865.0 | 482.1 | 320.5 | 0.471 Rank: 43/51 |
0.877 Rank: 1/51 |
0.42927 (±0.00000) Rank: 20/51 |
0.55738 (±0.00000) Rank: 19/51 |
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 dataset / Multiview track
mAA at 10 degrees: N/A (±0.00000 over 1 run(s) / ±0.13313 over 9 scenes)
Rank (per category): 24 (of 51)
| Scene | Features | Matches (input) |
RegistrationRatio (%) | Number of Landmarks |
Track Length | ATE | mAA(50) | mAA(100) |
| BM | 7969.3 | 831.36 | 99.93 Rank: 12/51 |
4220.06 Rank: 19/51 |
4.722 Rank: 37/51 |
0.48906 Rank: 32/51 |
0.47829 (±0.00000) Rank: 28/51 |
0.62473 (±0.00000) Rank: 27/51 |
| FCS | 7876.6 | 662.86 | 98.28 Rank: 18/51 |
4743.68 Rank: 28/51 |
4.266 Rank: 35/51 |
0.25021 Rank: 23/51 |
0.75043 (±0.00000) Rank: 18/51 |
0.80631 (±0.00000) Rank: 19/51 |
| LMS | 7773.0 | 654.93 | 99.40 Rank: 14/51 |
4356.13 Rank: 19/51 |
4.274 Rank: 35/51 |
0.27686 Rank: 7/51 |
0.85166 (±0.00000) Rank: 7/51 |
0.90530 (±0.00000) Rank: 5/51 |
| LB | 7958.1 | 639.74 | 98.03 Rank: 24/51 |
3710.81 Rank: 24/51 |
4.357 Rank: 37/51 |
0.48125 Rank: 21/51 |
0.69277 (±0.00000) Rank: 27/51 |
0.78280 (±0.00000) Rank: 27/51 |
| MC | 7828.2 | 738.14 | 99.81 Rank: 26/51 |
4418.89 Rank: 32/51 |
4.536 Rank: 38/51 |
0.40640 Rank: 38/51 |
0.51422 (±0.00000) Rank: 36/51 |
0.65594 (±0.00000) Rank: 36/51 |
| MR | 7795.9 | 699.05 | 95.01 Rank: 21/51 |
4807.05 Rank: 29/51 |
4.123 Rank: 39/51 |
0.48528 Rank: 19/51 |
0.46407 (±0.00000) Rank: 18/51 |
0.57517 (±0.00000) Rank: 18/51 |
| PSM | 7811.8 | 439.09 | 97.20 Rank: 25/51 |
4400.74 Rank: 27/51 |
3.237 Rank: 37/51 |
0.55776 Rank: 31/51 |
0.41638 (±0.00000) Rank: 33/51 |
0.49997 (±0.00000) Rank: 33/51 |
| SF | 7900.7 | 864.95 | 99.68 Rank: 13/51 |
5965.81 Rank: 27/51 |
4.436 Rank: 34/51 |
0.30046 Rank: 20/51 |
0.76165 (±0.00000) Rank: 18/51 |
0.84577 (±0.00000) Rank: 17/51 |
| SPC | 7871.1 | 817.23 | 100.00 Rank: 1/51 |
5184.35 Rank: 19/51 |
4.537 Rank: 30/51 |
0.43112 Rank: 13/51 |
0.75515 (±0.00000) Rank: 6/51 |
0.84577 (±0.00000) Rank: 5/51 |
| Avg | 7865.0 | 705.26 | 98.59 Rank: 21/51 |
4645.28 Rank: 24/51 |
4.276 Rank: 35/51 |
0.40871 Rank: 25/51 |
0.63162 (±0.00000) Rank: 22/51 |
0.72686 (±0.00000) Rank: 24/51 |
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 10 images. 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 —
Prague Parks dataset / Stereo track
mAA at 10 degrees: 0.59773 (±0.00000 over 1 run(s) / ±0.04042 over 3 scenes)
Rank (per category): 25 (of 51)
| Scene | Features | Matches (raw) |
Matches (final) |
Rep. @ 3 px. | MS @ 3 px. | mAA(5o) | mAA(10o) |
| Lizard | 8000.0 | 240.6 | 225.0 | 0.121 Rank: 20/51 |
0.007 Rank: 38/51 |
0.42080 (±0.00000) Rank: 21/51 |
0.54313 (±0.00000) Rank: 22/51 |
| Pond | 8000.0 | 349.7 | 299.6 | 0.152 Rank: 18/51 |
0.057 Rank: 7/51 |
0.51102 (±0.00000) Rank: 20/51 |
0.63968 (±0.00000) Rank: 24/51 |
| Tree | 8000.0 | 175.7 | 153.3 | 0.106 Rank: 27/51 |
0.019 Rank: 30/51 |
0.50728 (±0.00000) Rank: 27/51 |
0.61038 (±0.00000) Rank: 26/51 |
| Avg | 8000.0 | 255.3 | 226.0 | 0.126 Rank: 18/51 |
0.027 Rank: 24/51 |
0.47970 (±0.00000) Rank: 23/51 |
0.59773 (±0.00000) Rank: 25/51 |
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.
— Lizard —
— Pond —
— Tree —
Prague Parks dataset / Multiview track
mAA at 10 degrees: N/A (±0.00000 over 1 run(s) / ±0.14800 over 3 scenes)
Rank (per category): 23 (of 51)
| Scene | Features | Matches (input) |
RegistrationRatio (%) | Number of Landmarks |
Track Length | ATE | mAA(50) | mAA(100) |
| Lizard | 8000.0 | 382.87 | 84.00 Rank: 23/51 |
962.65 Rank: 32/51 |
3.144 Rank: 29/51 |
13.75632 Rank: 9/51 |
0.56724 (±0.00000) Rank: 12/51 |
0.64787 (±0.00000) Rank: 16/51 |
| Pond | 8000.0 | 614.45 | 69.13 Rank: 23/51 |
1326.05 Rank: 32/51 |
2.841 Rank: 29/51 |
0.42698 Rank: 15/51 |
0.25042 (±0.00000) Rank: 28/51 |
0.28536 (±0.00000) Rank: 27/51 |
| Tree | 8000.0 | 375.94 | 70.63 Rank: 28/51 |
669.72 Rank: 37/51 |
2.822 Rank: 32/51 |
5.40654 Rank: 5/51 |
0.43222 (±0.00000) Rank: 25/51 |
0.46767 (±0.00000) Rank: 23/51 |
| Avg | 8000.0 | 457.75 | 74.59 Rank: 27/51 |
986.14 Rank: 32/51 |
2.935 Rank: 30/51 |
6.52995 Rank: 7/51 |
0.41663 (±0.00000) Rank: 23/51 |
0.46696 (±0.00000) Rank: 23/51 |
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 10 images. Keypoints are drawn in blue if they are part of the model, and in red otherwise.
— Lizard —
— Pond —
— Tree —
Google Urban dataset / Stereo track
mAA at 10 degrees: 0.30094 (±0.00000 over 1 run(s) / ±0.12480 over 17 scenes)
Rank (per category): 18 (of 51)
| Scene | Features | Matches (raw) |
Matches (final) |
mAA(5o) | mAA(10o) |
| Amsterdam (AMS) | 6126.2 | 183.6 | 136.5 | 0.18304 (±0.00000) Rank: 20/51 |
0.29598 (±0.00000) Rank: 21/51 |
| Bangkok (BGK) | 6403.9 | 168.0 | 122.7 | 0.05856 (±0.00000) Rank: 19/51 |
0.16476 (±0.00000) Rank: 17/51 |
| Barcelona (BCN) | 5657.4 | 126.0 | 90.8 | 0.11136 (±0.00000) Rank: 18/51 |
0.17546 (±0.00000) Rank: 21/51 |
| Buenos Aires (BAR) | 6586.7 | 192.0 | 142.8 | 0.17534 (±0.00000) Rank: 20/51 |
0.30137 (±0.00000) Rank: 18/51 |
| Cambridge (CAM) | 6200.9 | 116.2 | 76.6 | 0.21027 (±0.00000) Rank: 20/51 |
0.32158 (±0.00000) Rank: 22/51 |
| Cannes (CAN) | 5862.6 | 169.9 | 106.2 | 0.35484 (±0.00000) Rank: 19/51 |
0.44608 (±0.00000) Rank: 18/51 |
| Chicago (CHI) | 5709.0 | 62.9 | 33.9 | 0.07801 (±0.00000) Rank: 20/51 |
0.13490 (±0.00000) Rank: 20/51 |
| Helsinki (HEL) | 6653.0 | 254.4 | 171.0 | 0.34845 (±0.00000) Rank: 17/51 |
0.54124 (±0.00000) Rank: 18/51 |
| Madrid (MAD) | 7031.3 | 125.8 | 87.2 | 0.06930 (±0.00000) Rank: 12/51 |
0.16292 (±0.00000) Rank: 12/51 |
| Mountain View (MTV) | 6280.8 | 105.2 | 67.3 | 0.14180 (±0.00000) Rank: 19/51 |
0.23555 (±0.00000) Rank: 18/51 |
| New Orleans (NOR) | 6235.4 | 126.2 | 80.9 | 0.19763 (±0.00000) Rank: 19/51 |
0.26213 (±0.00000) Rank: 19/51 |
| San Francisco (SF) | 6356.8 | 145.7 | 103.9 | 0.09449 (±0.00000) Rank: 22/51 |
0.20525 (±0.00000) Rank: 22/51 |
| Singapore (SG) | 7303.6 | 168.5 | 101.6 | 0.25190 (±0.00000) Rank: 14/51 |
0.35791 (±0.00000) Rank: 14/51 |
| Sydney (SYD) | 6676.7 | 148.1 | 93.7 | 0.22445 (±0.00000) Rank: 19/51 |
0.35459 (±0.00000) Rank: 18/51 |
| Tokyo (TOK) | 6592.6 | 189.0 | 109.2 | 0.45098 (±0.00000) Rank: 18/51 |
0.58676 (±0.00000) Rank: 18/51 |
| Toronto (TOR) | 7153.6 | 102.0 | 57.6 | 0.14560 (±0.00000) Rank: 17/51 |
0.25880 (±0.00000) Rank: 18/51 |
| Zurich (ZRH) | 6010.7 | 161.4 | 117.3 | 0.23780 (±0.00000) Rank: 16/51 |
0.31065 (±0.00000) Rank: 17/51 |
| Average (Avg) | 6402.4 | 149.7 | 100.0 | 0.19611 (±0.00000) Rank: 18/51 |
0.30094 (±0.00000) Rank: 18/51 |
We show the inliers that survive the robust estimation loop (i.e. RANSAC), or those supplied with the submission if using custom matches. As the dataset does not contain depth, we color them based on the symmetric epopilar distance on normalized coordinates: from 0 (green) to 2e-4 (yellow), and in red above it. Note that these occasionally contain false positives.
— Amsterdam —
— Bangkok —
— Barcelona —
— Buenos Aires —
— Cambridge —
— Cannes —
— Chicago —
— Helsinki —
— Madrid —
— Mountain View —
— New Orleans —
— San Francisco —
— Singapore —
— Sydney —
— Tokyo —
— Toronto —
— Zurich —
Google Urban dataset / Multiview track
mAA at 10 degrees: N/A (±0.00000 over 1 run(s) / ±0.13167 over 17 scenes)
Rank (per category): 21 (of 51)
| Scene | Features | Matches (input) |
RegistrationRatio (%) | Number of Landmarks |
Track Length | ATE | mAA(50) | mAA(100) |
| AMS | 6126.2 | 245.40 | 78.46 Rank: 23/51 |
1248.16 Rank: 25/51 |
3.004 Rank: 25/51 |
22.21845 Rank: 12/51 |
0.16999 (±0.00000) Rank: 24/51 |
0.24821 (±0.00000) Rank: 24/51 |
| BGK | 6403.9 | 260.95 | 90.93 Rank: 23/51 |
1717.49 Rank: 22/51 |
3.090 Rank: 26/51 |
13.02547 Rank: 13/51 |
0.03299 (±0.00000) Rank: 4/51 |
0.08865 (±0.00000) Rank: 8/51 |
| BCN | 5657.4 | 187.97 | 72.33 Rank: 25/51 |
659.34 Rank: 24/51 |
3.000 Rank: 27/51 |
21.75118 Rank: 3/51 |
0.09413 (±0.00000) Rank: 17/51 |
0.14120 (±0.00000) Rank: 18/51 |
| BAR | 6586.7 | 243.06 | 64.29 Rank: 22/51 |
728.43 Rank: 24/51 |
2.819 Rank: 24/51 |
9.20875 Rank: 5/51 |
0.05899 (±0.00000) Rank: 19/51 |
0.10647 (±0.00000) Rank: 19/51 |
| CAM | 6200.9 | 204.67 | 61.73 Rank: 25/51 |
760.09 Rank: 25/51 |
2.665 Rank: 28/51 |
14.90405 Rank: 5/51 |
0.10581 (±0.00000) Rank: 20/51 |
0.14641 (±0.00000) Rank: 21/51 |
| CAN | 5862.6 | 229.85 | 47.47 Rank: 26/51 |
268.48 Rank: 27/51 |
2.592 Rank: 28/51 |
6.92789 Rank: 2/51 |
0.01823 (±0.00000) Rank: 22/51 |
0.02928 (±0.00000) Rank: 24/51 |
| CHI | 5709.0 | 163.84 | 73.19 Rank: 21/51 |
460.62 Rank: 19/51 |
3.138 Rank: 24/51 |
43.92775 Rank: 12/51 |
0.01792 (±0.00000) Rank: 21/51 |
0.03854 (±0.00000) Rank: 22/51 |
| HEL | 6653.0 | 358.46 | 91.95 Rank: 13/51 |
2529.72 Rank: 18/51 |
2.987 Rank: 27/51 |
12.50885 Rank: 12/51 |
0.28390 (±0.00000) Rank: 11/51 |
0.44680 (±0.00000) Rank: 4/51 |
| MAD | 7031.3 | 241.59 | 80.77 Rank: 18/51 |
1148.30 Rank: 20/51 |
2.924 Rank: 26/51 |
6.00325 Rank: 21/51 |
0.03818 (±0.00000) Rank: 18/51 |
0.09648 (±0.00000) Rank: 15/51 |
| MTV | 6280.8 | 211.97 | 77.50 Rank: 24/51 |
917.70 Rank: 22/51 |
2.955 Rank: 24/51 |
12.52790 Rank: 5/51 |
0.11161 (±0.00000) Rank: 22/51 |
0.18630 (±0.00000) Rank: 23/51 |
| NOR | 6235.4 | 222.94 | 50.99 Rank: 24/51 |
412.71 Rank: 24/51 |
2.610 Rank: 25/51 |
5.23375 Rank: 4/51 |
0.05048 (±0.00000) Rank: 24/51 |
0.07447 (±0.00000) Rank: 25/51 |
| SF | 6356.8 | 224.72 | 73.23 Rank: 23/51 |
859.91 Rank: 24/51 |
2.883 Rank: 24/51 |
13.56619 Rank: 9/51 |
0.06437 (±0.00000) Rank: 24/51 |
0.12017 (±0.00000) Rank: 24/51 |
| SG | 7303.6 | 279.64 | 64.50 Rank: 18/51 |
786.93 Rank: 18/51 |
2.772 Rank: 26/51 |
14.13982 Rank: 11/51 |
0.09694 (±0.00000) Rank: 16/51 |
0.14229 (±0.00000) Rank: 15/51 |
| SYD | 6676.7 | 246.32 | 68.53 Rank: 23/51 |
746.72 Rank: 17/51 |
2.718 Rank: 26/51 |
9.92851 Rank: 6/51 |
0.10727 (±0.00000) Rank: 21/51 |
0.16468 (±0.00000) Rank: 21/51 |
| TOK | 6592.6 | 288.11 | 84.43 Rank: 16/51 |
1848.10 Rank: 16/51 |
2.934 Rank: 26/51 |
12.34467 Rank: 16/51 |
0.47213 (±0.00000) Rank: 12/51 |
0.54159 (±0.00000) Rank: 13/51 |
| TOR | 7153.6 | 224.85 | 64.11 Rank: 25/51 |
678.89 Rank: 18/51 |
2.596 Rank: 28/51 |
11.99946 Rank: 11/51 |
0.06508 (±0.00000) Rank: 19/51 |
0.12020 (±0.00000) Rank: 19/51 |
| ZRH | 6010.7 | 222.85 | 73.62 Rank: 19/51 |
1094.62 Rank: 23/51 |
2.755 Rank: 27/51 |
15.56291 Rank: 11/51 |
0.05556 (±0.00000) Rank: 16/51 |
0.10077 (±0.00000) Rank: 18/51 |
| Avg | 6402.4 | 238.66 | 71.65 Rank: 23/51 |
992.13 Rank: 21/51 |
2.849 Rank: 26/51 |
14.45758 Rank: 7/51 |
0.10844 (±0.00000) Rank: 21/51 |
0.16427 (±0.00000) Rank: 21/51 |
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 10 images. Keypoints are drawn in blue if they are part of the model, and in red otherwise.
— Amsterdam —
— Bangkok —
— Barcelona —
— Buenos Aires —
— Cambridge —
— Cannes —
— Chicago —
— Helsinki —
— Madrid —
— Mountain View —
— New Orleans —
— San Francisco —
— Singapore —
— Sydney —
— Tokyo —
— Toronto —
— Zurich —
