Boxmot

The goal with this release is to add different possibilities for ReID models. A lot has happened in the field since DeepSORT was first release, this is an attempt to keep up with the latest advancements in ReID methods.

Important updates
- Enable Yolov5 model ensembling
- Update track.py to comply with the new yolov5 standards
- Implementation of Lambda as per Eq(5) in the paper, based on https://github.com/michael-camilleri/deep_sort
- Added different ReID model options (https://kaiyangzhou.github.io/deep-person-reid/MODEL_ZOO)

Bug fixes
- Limit high performance libraries threads to 1 to avoid that the tracker uses all the CPUs (https://github.com/mikel-brostrom/Yolov5_DeepSort_Pytorch/issues/48)
- Default half precision inference to false for visualization on windows (https://github.com/mikel-brostrom/Yolov5_DeepSort_Pytorch/issues/206)
- Fix MOT index off by one in txt files (https://github.com/mikel-brostrom/Yolov5_DeepSort_Pytorch/issues/217)


MOT16 Train evaluation results

Relevant changed/used hparams: imgz 640, standard DeepSORT ReIDmodel. Notice that
none of the models used during the evaluation has ever seen any of the MOT16 data.


CLEAR: osnet_ain_x1_0_yolov5_lambda02-pedestrianMOTA MOTP MODA CLR_Re CLR_Pr MTR PTR MLR sMOTA CLR_TP CLR_FN CLR_FP IDSW MT PT ML Frag
MOT16-02 33.472 78.434 33.943 36.191 94.15 16.667 37.037 46.296 25.666 6454 11379 401 84 9 20 25 265
MOT16-04 63.852 76.516 64.054 71.359 90.714 40.964 42.169 16.867 47.094 33936 13621 3474 96 34 35 14 534
MOT16-05 58.199 78.433 59.226 68.568 88.008 27.2 57.6 15.2 43.411 4675 2143 637 70 34 72 19 182
MOT16-09 62.203 83.786 63.002 74.986 86.22 48 48 4 50.045 3942 1315 630 42 12 12 1 115
MOT16-10 53.613 77.071 54.092 57.761 94.027 25.926 48.148 25.926 40.369 7115 5203 452 59 14 26 14 386
MOT16-11 66.045 85.201 66.318 77.556 87.343 50.725 36.232 13.043 54.568 7115 2059 1031 25 35 25 9 126
MOT16-13 40.367 75.18 40.795 44.734 91.907 16.822 45.794 37.383 29.264 5122 6328 451 49 18 49 40 282
COMBINED 55.122 78.109 55.506 61.915 90.62 30.174 46.228 23.598 41.567 68359 42048 7076 425 156 239 122 1890

Identity: osnet_ain_x1_0_yolov5_lambda02-pedestrianIDF1 IDR IDP IDTP IDFN IDFP
MOT16-02 39.015 27.006 70.255 4816 13017 2039
MOT16-04 65.92 58.887 74.86 28005 19552 9405
MOT16-05 68.887 61.279 78.652 4178 2640 1134
MOT16-09 57.422 53.681 61.724 2822 2435 1750
MOT16-10 58.597 47.297 76.992 5826 6492 1741
MOT16-11 62.009 58.535 65.922 5370 3804 2776
MOT16-13 52.611 39.109 80.352 4478 6972 1095
COMBINED 59.723 50.264 73.567 55495 54912 19940

Count: osnet_ain_x1_0_yolov5_lambda02-pedestrianDets GT_Dets IDs GT_IDs
MOT16-02 6855 17833 110 54
MOT16-04 37410 47557 155 83
MOT16-05 5312 6818 158 125
MOT16-09 4572 5257 50 25
MOT16-10 7567 12318 93 54
MOT16-11 8146 9174 125 69
MOT16-13 5573 11450 98 107
COMBINED 75435 110407 789 517

- [DeepSort](https://drive.google.com/drive/folders/1xhG0kRH1EX5B9_Iz8gQJb7UNnn_riXi6)

The goal with this release is to automate the whole evaluation process using the official MOTXX evaluation data and tools.

Major changes

- Added colab notebook
- Added bash script for automatically handling all the MOT16 evaluation process (data download, wights download, video generation and placing in right folder...)
- Update track.py to comply with the new yolov5 standards
- Added id, class and confidence to plotted bboxes
- Added LICENSE

Bug fix

- Fix bad initial kf predictions for new objects in the field of view according to https://github.com/mikel-brostrom/Yolov5_DeepSort_Pytorch/issues/166
- Fix img input sizes bug according to https://github.com/mikel-brostrom/Yolov5_DeepSort_Pytorch/issues/174
- Class updated for each track after each detection instead of only in the initialization phase. This led to wrong class ID being displayed for each bbox


MOT16 train evaluation


CLEAR: ch_yolov5m_deep_sort-pedestrianMOTA MOTP MODA CLR_Re CLR_Pr MTR PTR MLR sMOTA CLR_TP CLR_FN CLR_FP IDSW MT PT ML Frag
MOT16-02 33.887 77.114 34.397 38.109 91.124 16.667 38.889 44.444 25.165 6796 11037 662 91 9 21 24 198
MOT16-04 63.831 76.326 63.997 72.149 89.848 40.964 39.759 19.277 46.75 34312 13245 3877 79 34 33 16 369
MOT16-05 56.307 76.523 57.334 71.414 83.531 40 49.6 10.4 39.541 4869 1949 960 70 50 62 13 150
MOT16-09 62.507 81.999 63.401 77.268 84.784 52 40 8 48.598 4062 1195 729 47 13 10 2 70
MOT16-10 52.703 75.324 53.239 59.685 90.253 27.778 46.296 25.926 37.975 7352 4966 794 66 15 25 14 272
MOT16-11 64.138 84.251 64.465 79.191 84.32 50.725 34.783 14.493 51.666 7265 1909 1351 30 35 24 10 83
MOT16-13 30.332 68.89 30.847 41.956 79.065 9.3458 52.336 38.318 17.279 4804 6646 1272 59 10 56 41 281
COMBINED 53.776 76.957 54.177 62.913 87.807 32.108 44.681 23.211 39.28 69460 40947 9645 442 166 231 120 1423

Identity: ch_yolov5m_deep_sort-pedestrianIDF1 IDR IDP IDTP IDFN IDFP
MOT16-02 36.361 25.784 61.652 4598 13235 2860
MOT16-04 67.341 60.708 75.6 28871 18686 9318
MOT16-05 39.583 36.712 42.94 2503 4315 3326
MOT16-09 50.378 48.145 52.828 2531 2726 2260
MOT16-10 54.251 45.064 68.144 5551 6767 2595
MOT16-11 47.768 46.316 49.315 4249 4925 4367
MOT16-13 39.393 30.148 56.814 3452 7998 2624
COMBINED 54.619 46.877 65.426 51755 58652 27350

Count: ch_yolov5m_deep_sort-pedestrianDets GT_Dets IDs GT_IDs
MOT16-02 7458 17833 50 54
MOT16-04 38189 47557 99 83
MOT16-05 5829 6818 42 125
MOT16-09 4791 5257 21 25
MOT16-10 8146 12318 46 54
MOT16-11 8616 9174 49 69
MOT16-13 6076 11450 54 107
COMBINED 79105 110407 361 517




v.2.0
The goal with this release is to create a CI pipeline for `track.py`. Automatic weight download for DeepSORT.

Important updates

- MOT16 Evaluation based on [73](https://github.com/mikel-brostrom/Yolov5_DeepSort_Pytorch/issues/73)
- Adapted track script to new yolov5 v5.0 standards
- README update explaining how to track different classes
- CI pipeline for testing CPU inference added
- Automatic weight downloading

The goal with this release is to make a 2 stage tracker based on Yolov5 publicly available for the first time ever (according to my personal search done on github 😅)

Major updates

- Basic tracking working: Yolov5 passes detections to DeepSORT which handles the tracking .
- Updated tracker when no detections. Based on https://github.com/mikel-brostrom/Yolov5_DeepSort_Pytorch/issues/21
- Adapted track script to new yolov5 v4.0 standards

Bug fixes

- PyTorch 1.7 compatibility update

Models