UNI for segmentation

[Xujingkk]

Hi, Richard J. Chen,
Thank you for your great work on UNI!
From the paper it appears that UNI was used as a backbone with ViT Adapter (for including hierarchical feaures) and Mask2Former head for segmentation. However, I don't find this code base very straightforward to use, so would appreciate more insights in how to include UNI as backbone in the ViT-Adapter infrastructure.
How can I connect the timm checkpoint as a backbone together with the mask2former segmentation head? When trying to load your weights into the ViT Adapter, I receive several errors.
Any help is appreciated.