siyueyu / dcfm Goto Github PK

Thanks for the great job!
When are you going to push your codes?

请问更换backbone后为什么梯度值为None？requires_grad是True，更换后没办法训练了，精度一直在0.44

How to regulate model training? Do you use any validation dataset to ensure that the trained model does not overfits?

Hi, I am really interested in your work.

While studying your code, I got a question in here:

I don't really get why that line is necessary.

I would appreciate if you give me any insights!

Thank you

Thank for your nice work. I would like to know how to visualize the response map, like figure 1.Thank you very much.

Thanks for your work.
But the SC Loss seems not to converge based on the related code. Can the code available achieve comparable performance to that in paper?

It seems ```python train.py ''' cannot achieve the performance of paper. It there any other trick?