Upon multiple attepts, including changing the number of layers in the ResNet, the hyperparameters, the number of epochs, the number of C_1, etc, we finally choose this python script as our code.
The pt file is the corresponding file containing the training weights of this architechture.
The training time is relatively fast, normally around an hour, so it is quite easy to reproduce.
Note: remember to change the logger filename or create a directory named 'log' before running the script.
The result is shown below.
