I have a query regarding the training process in STEGO. In DINO, the training approach involves using a fixed backbone and training only the interchangeable head. I'm wondering if the same principle applies to STEGO. Specifically, during training on custom dataset, are the ViT-Base / ViT-Small components frozen, similar to how the backbone is treated in DINO?
Hi,
I have a query regarding the training process in STEGO. In DINO, the training approach involves using a fixed backbone and training only the interchangeable head. I'm wondering if the same principle applies to STEGO. Specifically, during training on custom dataset, are the ViT-Base / ViT-Small components frozen, similar to how the backbone is treated in DINO?
Thanks!