Y-Mona/DeepSORT_Pytorch_VeRi

using pythorch , deepsort and veri datasets to train your own tracking model

DeepSORT_Pytorch_VeRi

using pythorch , deepsort and veri datasets to train your own tracking model

datasets

(remember to change the path of your datasets and other setting in the .py)

1. cd deepsort/deep/data
2. put your VeRi in it
3. cd ../
4. use getVal.py to get val imgs
5. cd data
6. use move2folder.py to get the satisfactory structure(train & val)
7. this movement may have some problems,I did not modify it.you only need to move it manually to val or train.
8. the structure:
   data
   —train
   ——0001
   ———0001xxx.jpg
   —val
   ——0001
   ———0001xxx.jpg

my modification

DeepSORT_Pytorch_VeRi/deep_sort/deep/train.py

Line 30 in cca50f1

test_dir = os.path.join(root,"val")

change the test datasets

DeepSORT_Pytorch_VeRi/deep_sort/deep/train.py

Line 32 in cca50f1

torchvision.transforms.Resize((128,64)),

change the randomcrop to resize

train

python train.py

track

(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)

1. cd detector/YOLOv3
2. put in your cfg & weight
3. modify the input of detector.py (cfg,weight,names)
4. modify the parameters of yolov3_deepsort.py(.yaml)
5. modify the yaml you use
6. modify 'mask' of yolov3_deepsort.py(0 start,the line of the class you car tracking in xx.names)
7. python yolov3_deepsort.py xxx.avi

result

reference

https://github.com/ZQPei/deep_sort_pytorch https://blog.csdn.net/weixin_40194996/article/details/104779138