Deep Image Clustering on ImageNet 100 (val)
96.61NMIProbing 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 | 96.61 | 96.16 | 92.73 | |
| Probing Sup. ViT-L/16Backbone=ViT-L/16, Pre-training strategy=Supervised, Protocol=Linear Probing2023.03 | 96.13 | 95.76 | 91.9 | |
| TEMI Sup. ViT-L/16Backbone=ViT-L/16, Pre-training strategy=Supervised, Protocol=Clustering2023.03 | 94.95 | 92.5 | 87.95 | |
| Probing DINO VIT-B/16Backbone=ViT-B/16, Pre-training strategy=Self-supervised (DINO), Protocol=Linear Probing2023.03 | 93.29 | 92.74 | 86.3 | |
| Probing MSN ViT-L/16Backbone=ViT-L/16, Pre-training strategy=Self-supervised (MSN), Protocol=Linear Probing2023.03 | 93 | 92.42 | 85.74 | |
| TEMI CLIP ViT-L/14Backbone=ViT-L/14, Pre-training strategy=Natural language supervised (CLIP), Protocol=Clustering2023.03 | 90.06 | 83.43 | 75.81 | |
| TEMI MSN VIT-L/16Backbone=ViT-L/16, Pre-training strategy=Self-supervised (MSN), Protocol=Clustering2023.03 | 88.53 | 82.86 | 74.08 | |
| TEMI DINO VIT-B/16Backbone=ViT-B/16, Pre-training strategy=Self-supervised (DINO), Protocol=Clustering2023.03 | 85.65 | 75.05 | 65.45 | |
| ProposBackbone=ResNet-50, Protocol=Clustering2023.03 | 83.5 | — | 63.5 | |
| SCANBackbone=ResNet-50, Protocol=Clustering2023.03 | 80.8 | 68.9 | 57.6 |