Skip to content
Mike edited this page Jun 5, 2021 · 13 revisions

Welcome to the Yolov5_DeepSort_Pytorch wiki!

Tutorials


  • Evaluation
  1. Download MOT16 data from:

https://motchallenge.net/data/MOT16/

under downloads. The structure of the downloaded data:


MOT16
├── train
│   ├── MOT16-02
│   │       ├── det
│   │       ├── gt
│   │       ├── img1
│   │       └── seqinfo.ini
│   ├── MOT16-04
│   ├── MOT16-05
│   ├── MOT16-09
│   ├── MOT16-10
│   ├── MOT16-11
│   └── MOT16-13
│
└── test
    ├── MOT16-02
    ├── MOT16-04
    ├── MOT16-05
    ├── MOT16-09
    ├── MOT16-10
    ├── MOT16-11
    └── MOT16-13

  1. Transform all images under all img1 under each MOT folder into videos by:

ffmpeg -framerate 25 -i %06d.jpg -c:v libx264 -profile:v high -crf 20 -pix_fmt yuv420p MOT16-13.mp4

  1. Then run:

python3 track.py --source /home/mikel.brostrom/Documents/TrackEval/data/MOT16/train/MOT16-XX/MOT16-XX.mp4 –save-txt

on all the videos and save the results under this cloned repo:

https://github.com/JonathonLuiten/TrackEval

by creating a folder to store the results for the different sequences generated by the yolov5_deep_sort algorithm like this:


trackeval
 ├── data
     ├── trackers
         ├── mot_challange
             ├── MOT16-train
                 ├── ch_yolov5m_deep_sort
                     ├── MOT16-02.txt
                     ├── MOT16-04.txt
                     ├── MOT16-05.txt
                     ├── MOT16-09.txt
                     ├── MOT16-10.txt
                     ├── MOT16-11.txt
                     ├── MOT16-13.txt

Then you can run the combined evaluation (on all sequences) by:

cd trackeval

python scripts/run_mot_challenge.py --BENCHMARK MOT16 --TRACKERS_TO_EVAL ch_yolov5m_deep_sort --METRICS CLEAR Identity --USE_PARALLEL False --NUM_PARALLEL_CORES 4

if you want to evaluate a single sequence you can use the --SEQ_INFO flag:

python scripts/run_mot_challenge.py --SEQ_INFO MOT16-04 --BENCHMARK MOT16 --TRACKERS_TO_EVAL humancrowd_yolov5_deep_sort --METRICS CLEAR Identity --USE_PARALLEL False --NUM_PARALLEL_CORES 1

NOTICE! That this evaluation is on the train dataset