yolov7-tracking

1. Environment

• python=3.7.0
• pytorch=1.7.0
• torchvision=0.8.0
• cudatoolkit=11.0

!pip install -r requirements.txt
!pip install motmetrics
!pip install cython_bbox
!pip install lap
!pip install pycocotools

2. Dataset

In this case, Using two different dataset for training.

• COCO dataset (only include car, bus and truck). train images 13016
• VisDrone dataset (only include car, bus and truck). train images 6169

3. Training

The model training follows yolov7

# train p5 models
python train.py --workers 8 --device 0 --epochs 50 --batch-size 32 --data data/coco_custom.yaml --img 640 640 --cfg cfg/training/yolov7-tiny.yaml --weights 'yolov7-tiny.pt' --name yolov7 --hyp data/hyp.scratch_custom_tiny.yaml

4. Tracking

• Creating four color area on hidden image to detect the color of vehicle center changing (to south and to north). The demo is shown as below.
• Hidden image (Left)
• main image (Right)

python tracker/tracker_custom.py --img_size 640 --tracker bytetrack --model_path your model weight --track_dataset Video  --dataset tracker/config_files/yolov7_track_custom.yaml

5. Demo