captum icon indicating copy to clipboard operation
captum copied to clipboard

Published 20 hours ago • verified publisher icon pytorch

What methods supports 3D convolutions

Open Zeev1988 opened this issue 1 year ago • 0 comments

Hi thank you for the amazing tool. I'm currently working with a pretrained video classification model and trying to use Captum to get some insights of the model and features. I tried to use LRP() to extract attributes, but it doesn't supports conv3d layers. I added this type to SUPPORTED_LAYERS_WITH_RULES and now it seems to work, although I'm not sure it really returns something meaningful right now.

I have 2 questions:

  • Why didn't you add 3d types to the SUPPORTED_LAYERS_WITH_RULES dict?
  • Is there different methods that are more suitable for videos models?

Zeev1988 avatar Dec 12 '23 15:12 Zeev1988