zju-vipa / protopformer Goto Github PK

hello, It's a nice work! But I don't understand the training process. Does it have the same training algorithm as ProtoPNet? or others? I don't really understand. I would appreciate it if you could answer this question.

Hello,

I just noticed that the default argument for --data_path in all entry files (main.py, etc.) is `'./datasets/cub200_cropped/'. In the ProtoPFormer paper (and supplemental), it is stated that images are not cropped, unlike ProtoPNet. Can you clarify whether you crop the CUB (and other dataset) images?

Thank you!

hello, sorry to bother you. It seems that main_visualize.py only can visualize local prototypes. So how to visualize global prototypes？

Hi!
We really like your work and would like to work on it. It would really helpful if you could make the pretrained protopformer model on CUB dataset available to public? As a student, we don't really have access to a lot of gpus to train.
Please consider, thank you!