Paper summary
Paper summary: DINOv3
DINOv3 scales self-supervised vision training and introduces Gram anchoring to stop dense feature maps from degrading during long training, producing a frozen backbone that beats specialised models across many tasks.
- Long self-supervised training used to make dense (per-patch) features worse; Gram anchoring fixes that.
- The features work well frozen, with no fine-tuning, across a broad range of vision tasks.