yolo_deepsort

MOT base on yolo3/yolo4+deepsort, different from the official use of numpy to implement sort, the sort here reimplemented with pytorch, so it running at GPU.

Mainly dependent

pytorch >= 1.3
torchvision >= 0.4.0
opencv-python >= 4.1

Quick start

Clone the repositorty

git clone https://github.com/GlassyWing/yolo_deepsort

Download weights of yolo3 or yolo4

cd weights/
wget https://pjreddie.com/media/files/yolov3.weights
wget https://pjreddie.com/media/files/yolov3-tiny.weights
wget https://drive.google.com/file/d/1cewMfusmPjYWbrnuJRuKhPMwRe_b9PaT/view

Download weights of DeepSort

# download ckpt.t7 from
https://drive.google.com/drive/folders/1xhG0kRH1EX5B9_Iz8gQJb7UNnn_riXi6 to this folder

Run example

python video_deepsort.py

If you do not want to run the tracker, set the parameter tracker to None:

video_detector = VideoDetector(...
                               tracker=None)

Replacing Re-ID

Please read the tutorial.

Example

Training

This library does not contain a feasible training program, please refer to the training:

yolo

deepsort

References

https://github.com/eriklindernoren/PyTorch-YOLOv3

https://github.com/ZQPei/deep_sort_pytorch