Skip to content

How to evaluate on custom tracking dataset

Mike edited this page Nov 8, 2022 · 11 revisions
  1. Create the GT folder under
Yolov5_StrongSORT_OSNet/val_utils/data/gt/mot_challenge/MOTCUSTOM-test

Following the standard MOT structure

MOTCUSTOM-test
├── MOTCUSTOM-01
│   ├── gt
│   │   └── gt.txt
│   └── seqinfo.ini
├── MOTCUSTOM-02
│   │   ├── gt
│   │   └── gt.txt
│   └── seqinfo.ini
.
.
.
└── MOTCUSTOM-0N

where seqinfo.ini contains:

[Sequence]
seqLength=<X>

fill in the sequence length by for example:

ffprobe -v error -select_streams v:0 -count_packets
        -show_entries stream=nb_read_packets -of csv=p=0 <your-video>
  1. Create tracking sequences folder under
Yolov5_StrongSORT_OSNet/val_utils/data/MOTCUSTOM/test

Following the standard MOT structure

test
├── MOTCUSTOM-01
│   └── MOTCUSTOM-01
│       └── vid.mp4  # you can also split the video into individual frames and place them here
├── MOTCUSTOM-02
│   └── MOTCUSTOM-02
│       └── vid.mp4  # you can also split the video into individual frames and place them here
.
.
.
└── MOTCUSTOM-0N
  1. Add the sequences that you dataset is comprised of in a .txt file
Yolov5_StrongSORT_OSNet/val_utils/data/gt/mot_challenge/seqmaps/MOTCUSTOM-test.txt

Add the names to the sequences in your dataset there:

name
MOTCUSTOM-01
MOTCUSTOM-02
...
MOTCUSTOM-0N
  1. Now everything is setup for custom dataset evaluation
python val.py --yolo-weights <my_people_weights> --benchmark MOTCUSTOM --split test --tracking-method strongsort
                                                                                                      ocsort
                                                                                                      bytetrack