Patch-level Representation Learning for Self-supervised Vision Transformers

Benchmark Model Rank Results
frozen-feature-label-propagation-on-davis-2017-valDINO + SelfPatch (ViT-S/16)–J&F: 62.7
image-classification-on-imagenetDINO + SelfPatch (ViT-S/16)–Top 1 Accuracy: 75.6%
object-detection-on-coco-val2017DINO + SelfPatch (ViT-S/16)–box AP: 42.1