Training mask2former head with ViT adapter for semantic segmentation

[ayushnangia]

Hi I was wondering about the code for training the whole mask2former head with ViT adapter with the DINOv2 backbone for my custom dataset. Is there any guide on this or where to get started? I have already tried this Linear probing of DINOv2 for semantic segmentation

Not great results and any idea about multiscale and improvement for linear would be great.

[antopost]

There is an official notebook that implements the ViT adapter but the notebook doesn't work.
However, there is a PR fixing the notebook... that I couldn't get to work either.
But maybe it will help you further.