CSTrack: Rethinking the competition between detection and ReID in Multi-Object Tracking
Video not available.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Rendering of new sequences is currently deactivated due to heavy load.
Short name:
CSTrack
Detector:
Private
Description:
CSTrack proposes a strong ReID based one-shot MOT framework. It includes a novel cross-correlation network that can effectively impel the separate branches to learn task-dependent representations, and a scale-aware attention network that learns discriminative embeddings to improve the ReID capability. This work also provides an analysis of the weak data association ability in one-shot MOT methods. Our improvements make the data association ability of our one-shot model is comparable to two-stage methods while running more faster.
Reference:
C. Liang, Z. Zhang, Y. Lu, X. Zhou, B. Li. Rethinking the competition between detection and ReID in Multi-Object Tracking. In arXiv:2010.12138 [cs], 2020.
Last submitted:
November 17, 2020 (4 years ago)
Published:
October 21, 2020 at 15:50:12 CET
Submissions:
2
Project page / code:
Open source:
Yes
Hardware:
RTX2080ti
Runtime:
15.8 Hz
Benchmark performance:
| Sequence | MOTA | IDF1 | HOTA | MT | ML | FP | FN | Rcll | Prcn | AssA | DetA | AssRe | AssPr | DetRe | DetPr | LocA | FAF | ID Sw. | Frag |
| MOT17 | 74.9 | 72.6 | 59.3 | 978 (41.5) | 411 (17.5) | 23,847 | 114,303 | 79.7 | 95.0 | 57.9 | 61.1 | 63.2 | 76.6 | 66.0 | 78.6 | 83.3 | 1.3 | 3,567 (44.7) | 7,668 (96.2) |
Detailed performance:
| Sequence | MOTA | IDF1 | HOTA | MT | ML | FP | FN | Rcll | Prcn | AssA | DetA | AssRe | AssPr | DetRe | DetPr | LocA | FAF | ID Sw. | Frag |
| MOT17-01-DPM | 57.1 | 63.0 | 50.9 | 8 | 8 | 89 | 2,645 | 59.0 | 97.7 | 54.4 | 47.7 | 58.5 | 81.2 | 49.6 | 82.1 | 84.2 | 0.2 | 35 | 103 |
| MOT17-01-FRCNN | 57.1 | 63.0 | 50.9 | 8 | 8 | 89 | 2,645 | 59.0 | 97.7 | 54.4 | 47.7 | 58.5 | 81.2 | 49.6 | 82.1 | 84.2 | 0.2 | 35 | 103 |
| MOT17-01-SDP | 57.1 | 63.0 | 50.9 | 8 | 8 | 89 | 2,645 | 59.0 | 97.7 | 54.4 | 47.7 | 58.5 | 81.2 | 49.6 | 82.1 | 84.2 | 0.2 | 35 | 103 |
| MOT17-03-DPM | 89.2 | 82.1 | 67.8 | 134 | 0 | 4,441 | 6,610 | 93.7 | 95.7 | 64.0 | 72.2 | 69.0 | 79.6 | 78.1 | 79.8 | 83.8 | 3.0 | 263 | 626 |
| MOT17-03-FRCNN | 89.2 | 82.1 | 67.8 | 134 | 0 | 4,441 | 6,610 | 93.7 | 95.7 | 64.0 | 72.2 | 69.0 | 79.6 | 78.1 | 79.8 | 83.8 | 3.0 | 263 | 626 |
| MOT17-03-SDP | 89.2 | 82.1 | 67.8 | 134 | 0 | 4,441 | 6,610 | 93.7 | 95.7 | 64.0 | 72.2 | 69.0 | 79.6 | 78.1 | 79.8 | 83.8 | 3.0 | 263 | 626 |
| MOT17-06-DPM | 61.6 | 63.9 | 51.2 | 79 | 54 | 402 | 3,959 | 66.4 | 95.1 | 51.3 | 51.2 | 58.9 | 73.0 | 54.7 | 78.4 | 83.2 | 0.3 | 168 | 286 |
| MOT17-06-FRCNN | 61.6 | 63.9 | 51.2 | 79 | 54 | 402 | 3,959 | 66.4 | 95.1 | 51.3 | 51.2 | 58.9 | 73.0 | 54.7 | 78.4 | 83.2 | 0.3 | 168 | 286 |
| MOT17-06-SDP | 61.6 | 63.9 | 51.2 | 79 | 54 | 402 | 3,959 | 66.4 | 95.1 | 51.3 | 51.2 | 58.9 | 73.0 | 54.7 | 78.4 | 83.2 | 0.3 | 168 | 286 |
| MOT17-07-DPM | 64.9 | 61.2 | 48.7 | 22 | 6 | 880 | 4,903 | 71.0 | 93.2 | 44.4 | 53.6 | 49.4 | 70.3 | 58.4 | 76.6 | 83.0 | 1.8 | 152 | 356 |
| MOT17-07-FRCNN | 64.9 | 61.2 | 48.7 | 22 | 6 | 880 | 4,903 | 71.0 | 93.2 | 44.4 | 53.6 | 49.4 | 70.3 | 58.4 | 76.6 | 83.0 | 1.8 | 152 | 356 |
| MOT17-07-SDP | 64.9 | 61.2 | 48.7 | 22 | 6 | 880 | 4,903 | 71.0 | 93.2 | 44.4 | 53.6 | 49.4 | 70.3 | 58.4 | 76.6 | 83.0 | 1.8 | 152 | 356 |
| MOT17-08-DPM | 53.0 | 45.6 | 39.6 | 25 | 9 | 985 | 8,590 | 59.3 | 92.7 | 36.0 | 44.7 | 43.0 | 59.0 | 48.3 | 75.5 | 82.1 | 1.6 | 351 | 548 |
| MOT17-08-FRCNN | 53.0 | 45.6 | 39.6 | 25 | 9 | 985 | 8,590 | 59.3 | 92.7 | 36.0 | 44.7 | 43.0 | 59.0 | 48.3 | 75.5 | 82.1 | 1.6 | 351 | 548 |
| MOT17-08-SDP | 53.0 | 45.6 | 39.6 | 25 | 9 | 985 | 8,590 | 59.3 | 92.7 | 36.0 | 44.7 | 43.0 | 59.0 | 48.3 | 75.5 | 82.1 | 1.6 | 351 | 548 |
| MOT17-12-DPM | 56.5 | 66.6 | 53.9 | 26 | 22 | 617 | 3,100 | 64.2 | 90.0 | 58.6 | 49.8 | 64.1 | 78.5 | 54.7 | 76.7 | 84.9 | 0.7 | 49 | 176 |
| MOT17-12-FRCNN | 56.5 | 66.6 | 53.9 | 26 | 22 | 617 | 3,100 | 64.2 | 90.0 | 58.6 | 49.8 | 64.1 | 78.5 | 54.7 | 76.7 | 84.9 | 0.7 | 49 | 176 |
| MOT17-12-SDP | 56.5 | 66.6 | 53.9 | 26 | 22 | 617 | 3,100 | 64.2 | 90.0 | 58.6 | 49.8 | 64.1 | 78.5 | 54.7 | 76.7 | 84.9 | 0.7 | 49 | 176 |
| MOT17-14-DPM | 51.3 | 61.2 | 42.9 | 32 | 38 | 535 | 8,294 | 55.1 | 95.0 | 46.2 | 40.3 | 49.8 | 74.1 | 42.7 | 73.6 | 79.5 | 0.7 | 171 | 461 |
| MOT17-14-FRCNN | 51.3 | 61.2 | 42.9 | 32 | 38 | 535 | 8,294 | 55.1 | 95.0 | 46.2 | 40.3 | 49.8 | 74.1 | 42.7 | 73.6 | 79.5 | 0.7 | 171 | 461 |
| MOT17-14-SDP | 51.3 | 61.2 | 42.9 | 32 | 38 | 535 | 8,294 | 55.1 | 95.0 | 46.2 | 40.3 | 49.8 | 74.1 | 42.7 | 73.6 | 79.5 | 0.7 | 171 | 461 |
Raw data:
