code issues

[Liuhao-128]

Hi, Hurtado, I am interesting in your work.
I have a questions about the code in approach.py. Specifically, I am not understand what is the meaning of input_feats=batch[2] .
The shape of input_feats is [batchsize, channel, 28, 28].
Could your tell me the answer, thanks a lot.

[JuliousHurtado]

Hi,
thank you for being interested in our work.

About your question.
If you check the function __getitem__ in the dataloaders of the different benchmarks, you can observed that it has 3 output.

• batch[0]: is the input of the model (images)
• batch[1]: is the corresponding label
• batch[2]: is the feature vector of a pre-trained model

This last output is nor really used in the basic method proposed, only in the experiment when we use a pre-trained model as an F function.