pytorch-image-models
pytorch-image-models copied to clipboard
[FEATURE] Modify SAM -> ViTDet
Segment Anything (SAM) uses the ViTDet backbone [paper]. The only difference is that SAM adds a 3 layer neck. This neck can be disabled through arguments. By changing this 1 option, ViTDet can be reproduced.
However, ViTDet is originally meant as a feature pyramid network. That is the reason for its window strategy. "Features_only" should be supported by ViTDet / SAM.