using pythorch , deepsort and veri datasets to train your own tracking model
(remember to change the path of your datasets and other setting in the .py)
- cd deepsort/deep/data
- put your VeRi in it
- cd ../
- use getVal.py to get val imgs
- cd data
- use move2folder.py to get the satisfactory structure(train & val)
- this movement may have some problems,I did not modify it.you only need to move it manually to val or train.
- the structure:
data
—train
——0001
———0001xxx.jpg
—val
——0001
———0001xxx.jpg
python train.py
(require .cfg & .weight which is trained by yolov3. yolov3-spp.cfg could not run. For details,see my repo https://github.com/Y-Mona/YOLOv3_Pytorch_UA-DETRAC)
- cd detector/YOLOv3
- put in your cfg & weight
- modify the input of detector.py (cfg,weight,names)
- modify the parameters of yolov3_deepsort.py(.yaml)
- modify the yaml you use
- modify 'mask' of yolov3_deepsort.py(0 start,the line of the class you car tracking in xx.names)
- python yolov3_deepsort.py xxx.avi
https://github.com/ZQPei/deep_sort_pytorch https://blog.csdn.net/weixin_40194996/article/details/104779138