JackWoo0831 / Yolov7-tracker

Yolo X, v7, v8 and several Multi-Object Tracker(SORT, DeepSORT, ByteTrack, BoT-SORT, etc.) in MOT17 and VisDrone2019 Dataset. It uses a unified style and integrated tracker for easy embedding in your own projects.
GNU General Public License v3.0
636 stars 88 forks source link

为什么这个SORT的MOTA反而更高呢 #102

Open LEIJUNMIUI opened 7 months ago

LEIJUNMIUI commented 7 months ago

在测试结果中,SORT的MOTA比其它算法高,速度反而比bytetrack慢

JackWoo0831 commented 7 months ago

我也很奇怪... 也许是评测代码的问题 你实际跑起来结果如何呢

LEIJUNMIUI commented 7 months ago

我还没测 测完了看看

SERBOLD commented 6 months ago

我也遇到了这个问题: sort HOTA: ./result-pedestrian HOTA DetA AssA DetRe DetPr AssRe AssPr LocA OWTA HOTA(0) LocA(0) HOTALocA(0) seq1_changes 61.253 61.801 60.856 72.722 68.205 65.23 75.78 79.254 66.51 84.48 74.327 62.791 COMBINED 61.253 61.801 60.856 72.722 68.205 65.23 75.78 79.254 66.51 84.48 74.327 62.791

CLEAR: ./result-pedestrian MOTA MOTP MODA CLR_Re CLR_Pr MTR PTR MLR sMOTA CLR_TP CLR_FN CLR_FP IDSW MT PT ML Frag seq1_changes 76.525 75.573 77.618 92.12 86.399 100 0 0 54.023 7248 620 1141 86 54 0 0 117 COMBINED 76.525 75.573 77.618 92.12 86.399 100 0 0 54.023 7248 620 1141 86 54 0 0 117

Identity: ./result-pedestrian IDF1 IDR IDP IDTP IDFN IDFP seq1_changes 82.5 85.231 79.938 6706 1162 1683 COMBINED 82.5 85.231 79.938 6706 1162 1683

Count: ./result-pedestrian Dets GT_Dets IDs GT_IDs seq1_changes 8389 7868 290 54 COMBINED 8389 7868 290 54

deepsort HOTA: ./result-pedestrian HOTA DetA AssA DetRe DetPr AssRe AssPr LocA OWTA HOTA(0) LocA(0) HOTALocA(0) seq1_changes 51.058 55.568 47.216 68.263 64.811 50.906 72.992 80.051 56.731 66.653 75.278 50.175 COMBINED 51.058 55.568 47.216 68.263 64.811 50.906 72.992 80.051 56.731 66.653 75.278 50.175

CLEAR: ./result-pedestrian MOTA MOTP MODA CLR_Re CLR_Pr MTR PTR MLR sMOTA CLR_TP CLR_FN CLR_FP IDSW MT PT ML Frag seq1_changes 62.265 76.928 65.264 85.295 80.982 96.296 3.7037 0 42.586 6711 1157 1576 236 52 2 0 633 COMBINED 62.265 76.928 65.264 85.295 80.982 96.296 3.7037 0 42.586 6711 1157 1576 236 52 2 0 633

Identity: ./result-pedestrian IDF1 IDR IDP IDTP IDFN IDFP seq1_changes 69.774 71.632 68.01 5636 2232 2651 COMBINED 69.774 71.632 68.01 5636 2232 2651

Count: ./result-pedestrian Dets GT_Dets IDs GT_IDs seq1_changes 8287 7868 356 54 COMBINED 8287 7868 356 54

测量的是柑橘,对比的数据集是我手动标的

chenshengyeah111 commented 3 months ago

在测试结果中,SORT的MOTA比其它算法高,速度反而比bytetrack慢 为什么我的结果特别低(SORT) HOTA: LPC_MOT-pedestrian HOTA DetA AssA DetRe DetPr AssRe AssPr LocA OWTA HOTA(0) LocA(0) HOTALocA(0) MOT20-01 38.337 48.78 30.375 50.99 85.191 47.233 41.134 87.389 39.284 43.993 84.201 37.042 MOT20-02 24.619 45.598 13.39 47.53 85.755 36.369 17.752 88.023 25.185 27.926 84.995 23.736 MOT20-03 22.071 40.779 12.038 43.933 77.367 36.829 14.664 84.852 22.958 26.144 80.125 20.948 MOT20-05 18.054 42.503 7.7468 45.278 80.252 28.756 10.729 85.794 18.68 21.205 80.817 17.137 COMBINED 20.705 42.531 10.185 45.313 80.278 32.456 13.438 85.886 21.425 24.193 81.264 19.66

CLEAR: LPC_MOT-pedestrian MOTA MOTP MODA CLR_Re CLR_Pr MTR PTR MLR sMOTA CLR_TP CLR_FN CLR_FP IDSW MT PT ML Frag MOT20-01 55.068 86.208 56.14 57.997 96.897 24.324 60.811 14.865 47.069 11524 8346 369 213 18 45 11 260 MOT20-02 51.469 87.062 52.292 53.859 97.174 25.185 62.593 12.222 44.5 83342 71400 2424 1274 68 169 33 1548 MOT20-03 44.295 83.439 45.48 51.133 90.046 14.815 56.268 28.917 35.827 160383 153275 17730 3719 104 395 203 4129 MOT20-05 46.822 84.608 48.354 52.387 92.853 19.076 57.998 22.926 38.758 338598 307746 26063 9904 223 678 268 10077 COMBINED 46.902 84.667 48.233 52.339 92.726 18.646 58.104 23.251 38.877 593847 540767 46586 15110 413 1287 515 16014

Identity: LPC_MOT-pedestrian IDF1 IDR IDP IDTP IDFN IDFP MOT20-01 40.758 32.577 54.427 6473 13397 5420 MOT20-02 20.562 15.98 28.831 24727 130015 61039 MOT20-03 19.394 15.204 26.773 47687 265971 130426 MOT20-05 14.713 11.507 20.395 74374 571970 290287 COMBINED 17.268 13.508 23.931 153261 981353 487172

JackWoo0831 commented 3 months ago

您好,我已经更新了一版代码,branch是v2,不然您试一下最新版的代码怎么样