Mask Probing on ImageNet (val)
51.8Jaccard SimilarityGroupViT
Evaluation Results
| Method | Links | |
|---|---|---|
| GroupViTarch=GroupViT, model=Ours, dataset (pre-training)=CC12M+YFCC, supervision=text2022.02 | 51.8 | |
| DINOarch=ViT, model=DINO, dataset (pre-training)=ImageNet, supervision=self2022.02 | 45.9 | |
| DINOarch=ViT, model=DINO, dataset (pre-training)=CC12M+YFCC, supervision=self2022.02 | 41.8 | |
| CLIParch=ViT, model=CLIP, dataset (pre-training)=CC12M+YFCC, supervision=text2022.02 | 28.6 | |
| MoCoarch=ViT, model=MoCo, dataset (pre-training)=ImageNet, supervision=self2022.02 | 28.2 | |
| DeiTarch=ViT, model=DeiT, dataset (pre-training)=ImageNet, supervision=class2022.02 | 24.6 | |
| ViTarch=ViT, model=Random, dataset (pre-training)=none, supervision=none2022.02 | 23.6 |