Firstly, what I have to say is: my framework was built on this repository, and the original author has done a very good job. On this basis, I added the functions of training custom datasets and adapting custom networks, and added the test code and visualized results to fix the bugs of the original evaluation code.

Basically, you can follow the instructions in this repository to configure the environment, but I added some other dependencies to requirements.txt. So in order to run these codes better, follow the steps below to install them.

conda create -n img-class python=3.8

conda install pytorch==1.10.1 torchvision==0.11.2 cudatoolkit=11.3

pip install -r requirements.txt