about the ckp epoch

[Vijayue]

Thanks a lot for your code and pre-trained model.

Now I want to continue training on your pretrained model, after loading the pretrained model, the epoch begins from the 12400, but the ckp name is 00000100-ckp.pth.tar, which means the ckp was generated after the 100 epochs? Do you have any idea about the issue? Thank you!

[zhengkw18]

I'm sorry this mismatch may confuse you. This is caused by the change of 'num_repeats' in 'train.py'. When I train the model, it is set to 1, so each epoch takes only a few minutes to facilitate faster debugging. When I release the code, it is set to 100, which is consistent with FOMM. In the current config, the checkpoint is roughly at the progress of 124 epoch.

[Vijayue]

OK! Thank you for your explanation, I got it!