search

yangyangyang127 / APE

[ICCV 2023] Code for "Not All Features Matter: Enhancing Few-shot CLIP with Adaptive Prior Refinement"
135 stars 12 forks source link

About the usage of APE-T #3

Open myt889 opened 1 year ago

myt889 commented 1 year ago

Hi, thank you for interesting work. I had one questions: If I wonder if APE-T is suitable for specified supervised classification task? If so, how can I adapt it to new task? thank you so much.

yangyangyang127 commented 1 year ago

Hi, thanks for your interest. In my humble opinion, APE-T can be used for other tasks but there may be two key concerns to using it for fully supervised tasks. 1) The channel selection module is more suitable for few-shot tasks, thus it may suffer a setback when more training samples are given. 2) According to my observation, the residual added to the textual features is more important than other residuals or weights. I hope this may be helpful for your experiments.

myt889 commented 1 year ago

Thank you so much, I will consider your opinion and make a try.