Version: 0.1
Support: [convolutional] [maxpool] [reorg] [route] [shortcut] [region] keywords in the network config file, yolov2 network
Requirement:
1 python3.5
2 pytorch0.4
3 torchvision
4 PIL/Pillow
5 opencv-python
Get start:
1 training: require a dataset config file and a network config file, refer to the dataset/coco.data and cfg/yolov2.cfg
python3 train.py --init=1(this command will orthogonal initialize all parameters in the network and training a new model)
if you want to visualize the training process, python3 -m visdom.server, the go to the http://localhost:8097 in your browser
2 python3 test.py
for more details, please see the source code
To do:
1 add the loss function of yolov3
2 add evalution function during training
3 speed up the loss function
4 add vedio detection demo