semantic segmentation #281
Description
I am working on semantic segmentation using the DPT as decoder. Under the same settings, DINOv3 performs worse than DINOv2. Is there a reason why DINOv3 is less effective for dense prediction tasks?
{{ message }}