STC_Tracker: Strong TransCenter-based Tracker
MOT20-04
Short name:
STC_Tracker
Detector:
Private
Description:
Transformer detector based on TransCenter (Yihung Xu et al.) with a Kalman filter motion model as in StrongSORT (Yunhao Du et al.) and BOT-SORT (Nir Aharon et al.) and with FastReid (Lingxiao He et al.) for additional association accuracy
Reference:
A. Galor, R. Orfaig, B. Bobrovsky. Strong-TransCenter: Improved Multi-Object Tracking based on Transformers with Dense Representations. In arXiv preprint arXiv:2210.13570, 2022.
Last submitted:
September 22, 2022 (3 years ago)
Published:
February 27, 2023 at 14:09:51 CET
Submissions:
1
Project page / code:
Open source:
Yes
Hardware:
CPU 2.1GHz, 1 core, GPU geforce gtx 1180 8GB
Runtime:
1.4 Hz
Benchmark performance:
| Sequence | MOTA | IDF1 | HOTA | MT | ML | FP | FN | Rcll | Prcn | AssA | DetA | AssRe | AssPr | DetRe | DetPr | LocA | FAF | ID Sw. | Frag |
| MOT20 | 73.0 | 67.5 | 56.3 | 832 (67.0) | 147 (11.8) | 30,215 | 107,701 | 79.2 | 93.1 | 53.1 | 59.8 | 59.4 | 73.4 | 65.5 | 77.1 | 83.3 | 6.7 | 2,011 (0.0) | 3,069 (0.0) |
Detailed performance:
| Sequence | MOTA | IDF1 | HOTA | MT | ML | FP | FN | Rcll | Prcn | AssA | DetA | AssRe | AssPr | DetRe | DetPr | LocA | FAF | ID Sw. | Frag |
| MOT20-04 | 84.3 | 76.5 | 63.6 | 578 | 19 | 19,537 | 22,690 | 91.7 | 92.8 | 59.0 | 68.7 | 65.9 | 76.0 | 76.3 | 77.2 | 83.5 | 9.4 | 710 | 1,149 |
| MOT20-06 | 59.1 | 53.6 | 44.9 | 113 | 71 | 5,250 | 48,412 | 63.5 | 94.1 | 41.5 | 48.7 | 46.7 | 68.3 | 52.2 | 77.4 | 83.0 | 5.2 | 607 | 949 |
| MOT20-07 | 78.3 | 69.3 | 58.8 | 86 | 3 | 2,264 | 4,665 | 85.9 | 92.6 | 53.9 | 64.8 | 59.4 | 74.2 | 72.0 | 77.7 | 84.1 | 3.9 | 242 | 288 |
| MOT20-08 | 54.1 | 52.3 | 42.5 | 55 | 54 | 3,164 | 31,934 | 58.8 | 93.5 | 41.1 | 44.3 | 46.3 | 67.4 | 47.6 | 75.7 | 82.3 | 3.9 | 452 | 683 |
Raw data:
