GeneratorNet forward pass

[ashok-arjun]

The forward pass of the GeneratorNet network and Net network (the feature extractor) multiply A_rebuild with self.s which is set to 10 and then passes it to the classifier.

Can you explain what this is? Why can't we pass it directly to the FC layer for classification?

[Yuxin-CV]

Hi, thanks for your interest in our work.

self.s corresponds to the learnable temperature \alpha described in our paper.
To my knowledge, the term "temperature" comes from [1], and is adopted by [2](which is also cited by us) in few-shot learning.
In a word, softmax w/ learnable temperature \alpha is easier to optimize.
You can refer to [1, 2] for more details.

[1] Distilling the Knowledge in a Neural Network.
[2] Low-shot learning with imprinted weights.

[ashok-arjun]

Thank you for the explanation! I'm clear now.

Thanks for releasing the code for this great work!