Xiaocc
hi, i think the term_width is the width of terminal, you can try to use command line to train the model or fix the term_width number.
Because in the test phase we use the 'net' to extract the features and do not need the weights of 'arcmargin'.
Sorry, I didn't understand your question too much, can you describe it in detail？
running_loss is just a variable that sums the total loss for each training period.
I use the pre-trained model. https://github.com/pytorch/examples/blob/5df464c46cf321ed1cc3df1e670358d7f5ae1887/imagenet/main.py#L198