Vision Transformer reconstruct image

🤗Transformers
#1

Dear,
from the output of a ViT how I can reconstruct an image of features like the output of a CNN?

I need this kind of representation for a network that uses that representation for the information information.

#2

Hello,
I need to obtain an output like the features images in the video of that image

attention_maps
attention_maps3816×968 872 KB

to use that image as input for another piece of the network.

#3

Hi,

See my notebook here: Transformers-Tutorials/Visualize_self_attention_of_DINO.ipynb at master · NielsRogge/Transformers-Tutorials · GitHub.