Swin_JDE: Joint Detection and Embedding Multi-Object Tracking Based on Swin-Transformer
MOT20-04
Short name:
Swin_JDE
Detector:
Private
Description:
Swin-JDE is a novel anchor-free joint detection and embedding (JDE) MOT method based on Transformer architecture, which provides excellent tracking performance in crowded scenes.
Reference:
C.-Y. Tsai, G.-Y. Shen and H. Nisar, "Swin-JDE: Joint Detection and Embedding Multi-Object Tracking Based on Swin-Transformer," Engineering Applications of Artificial Intelligence, Vol. 119, No. 105770, pp. 1-16, 2023.
Last submitted:
June 29, 2022 (2 years ago)
Published:
August 03, 2022 at 04:55:49 CET
Submissions:
1
Project page / code:
n/a
Open source:
No
Hardware:
3.6GHZ, 4 Cores
Runtime:
4.1 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 | 70.4 | 69.5 | 55.7 | 730 (58.8) | 167 (13.4) | 26,394 | 124,829 | 75.9 | 93.7 | 54.8 | 56.9 | 58.8 | 77.6 | 61.8 | 76.3 | 82.1 | 5.9 | 2,026 (0.0) | 3,165 (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 | 84.2 | 80.0 | 64.4 | 498 | 34 | 9,862 | 32,705 | 88.1 | 96.1 | 61.6 | 67.6 | 65.8 | 80.0 | 72.7 | 79.3 | 83.1 | 4.7 | 667 | 1,103 |
| MOT20-06 | 54.3 | 54.7 | 43.0 | 103 | 71 | 8,161 | 51,768 | 61.0 | 90.8 | 42.0 | 44.3 | 45.4 | 72.1 | 48.3 | 72.0 | 80.0 | 8.1 | 754 | 1,192 |
| MOT20-07 | 76.3 | 68.2 | 55.4 | 77 | 5 | 1,827 | 5,783 | 82.5 | 93.7 | 50.3 | 61.5 | 54.1 | 75.1 | 67.2 | 76.3 | 81.9 | 3.1 | 224 | 287 |
| MOT20-08 | 46.4 | 52.8 | 40.5 | 52 | 57 | 6,544 | 34,573 | 55.4 | 86.8 | 41.7 | 39.4 | 44.5 | 73.6 | 43.8 | 68.6 | 80.0 | 8.1 | 381 | 583 |
Raw data:
