The original logic was through CIFAR10 VGG calls train Main in classify.py firing The rate method, but the original main firing The rate method does not have training and testing code, it only outputs the network structure. My modifications have added training code and changed the loss calculation method to mse. However, there is a new problem, which is that as the loss decreases, the model seems to have not learned anything (the accuracy of classification has always been 10%/50%).
The original logic was through CIFAR10 VGG calls train Main in classify.py firing The rate method, but the original main firing The rate method does not have training and testing code, it only outputs the network structure. My modifications have added training code and changed the loss calculation method to mse. However, there is a new problem, which is that as the loss decreases, the model seems to have not learned anything (the accuracy of classification has always been 10%/50%).