Deep Image Clustering on ImageNet 200 (val)
95.09NMIProbing CLIP ViT-L/14
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Probing CLIP ViT-L/14Backbone=ViT-L/14, Pre-training strategy=Natural language supervised (CLIP), Protocol=Linear Probing2023.03 | 95.09 | 93.57 | 88 | |
| Probing Sup. ViT-L/16Backbone=ViT-L/16, Pre-training strategy=Supervised, Protocol=Linear Probing2023.03 | 95.07 | 93.6 | 88.02 | |
| TEMI Sup. ViT-L/16Backbone=ViT-L/16, Pre-training strategy=Supervised, Protocol=Clustering2023.03 | 93.94 | 90.37 | 84.05 | |
| Probing DINO VIT-B/16Backbone=ViT-B/16, Pre-training strategy=Self-supervised (DINO), Protocol=Linear Probing2023.03 | 91.64 | 89.48 | 80.61 | |
| Probing MSN ViT-L/16Backbone=ViT-L/16, Pre-training strategy=Self-supervised (MSN), Protocol=Linear Probing2023.03 | 91.36 | 89.02 | 79.88 | |
| TEMI CLIP ViT-L/14Backbone=ViT-L/14, Pre-training strategy=Natural language supervised (CLIP), Protocol=Clustering2023.03 | 88.39 | 77.76 | 69.41 | |
| TEMI MSN VIT-L/16Backbone=ViT-L/16, Pre-training strategy=Self-supervised (MSN), Protocol=Clustering2023.03 | 86.65 | 77.96 | 66.7 | |
| TEMI DINO VIT-B/16Backbone=ViT-B/16, Pre-training strategy=Self-supervised (DINO), Protocol=Clustering2023.03 | 85.2 | 73.12 | 62.13 | |
| ProposBackbone=ResNet-50, Protocol=Clustering2023.03 | 80.6 | — | 53.8 | |
| SCANBackbone=ResNet-50, Protocol=Clustering2023.03 | 77.2 | 58.1 | 47 |