SSF in ResBlocks

[meeselizabeth]

Hi, I am trying your SSF method by inserting it in the ResBlocks of my network after the norm and conv operations. I use a pretrained model, freeze its weights and just train the SSF params, however it scores very bad although the dataset the pretrained model is trained on and the new dataset are very similar, just different number of classes. I was wondering what you are doing with the heads of your pretrained models, in your code I found something like exclude_head='agc', what does that do?

Best