Can Vision Transformers be used to extract features, just like with VGG ?

I am interested in using this vision transformer in extracting features (https://huggingface.co/google/vit-base-patch16-224)

Since VIT is used in classification problems, I don't think that it will properly encode features like for example VAEs.

I want to use these features in image generation, something akin to super resolution.


1 Answer 1


Yes, of course they can be used to extract features, just like convolutional networks, even in supervised settings.

ViTs are not exclusive to classification, their intermediate layers can also be used as high quality features, just to show this, if your search in Google Scholar for "vision transformer transfer learning", at the time this question was written, there are 66700 papers.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .