Clustering on ImageNet (val)
69.2AMIMAE-CT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MAE-CTBackbone=ViT-H/16, Augmentation=minimal image augmentations, variant=MAE-CTmin2023.04 | 69.2 | 57.1 | 81.7 | 35.6 | |
| MAE-CTBackbone=ViT-L/16, Augmentation=extensive image augmentations, variant=MAE-CTaug2023.04 | 68 | 55.6 | 81.4 | 39.4 | |
| MAE-CTBackbone=ViT-H/16, Augmentation=extensive image augmentations, variant=MAE-CTaug2023.04 | 67.3 | 54.7 | 80.8 | 37.8 | |
| MAE-CTBackbone=ViT-L/16, Augmentation=minimal image augmentations, variant=MAE-CTmin2023.04 | 66.9 | 54.1 | 80.6 | 34.7 | |
| iBOTBackbone=ViT-L/16, Augmentation=extensive image augmentations2023.04 | 66.4 | 51.2 | 80.1 | 34.1 | |
| MSNBackbone=ViT-B/16, Augmentation=extensive image augmentations2023.04 | 65.9 | 52.7 | 79.4 | 28.8 | |
| iBOTBackbone=ViT-B/16, Augmentation=extensive image augmentations2023.04 | 64.6 | 49.2 | 78.7 | 31.6 | |
| DINOBackbone=ViT-B/16, Augmentation=extensive image augmentations2023.04 | 62.7 | 46.5 | 77.3 | 29 | |
| MSNBackbone=ViT-L/16, Augmentation=extensive image augmentations2023.04 | 62.7 | 43.6 | 75.6 | 18.6 | |
| TEMI DINOArchitecture=ViT-B/16, Number of clusters=25K, Beta factor=12023.03 | 59.9 | — | — | — | |
| MAE-CTBackbone=ViT-B/16, Augmentation=extensive image augmentations, variant=MAE-CTaug2023.04 | 58.8 | 45.3 | 75.7 | 28.6 | |
| TEMI MSNArchitecture=ViT-L/16, Number of clusters=25K, Beta factor=12023.03 | 58.8 | — | — | — | |
| Moco v3Backbone=ViT-B/16, Augmentation=extensive image augmentations2023.04 | 55.2 | 41.5 | 72.4 | 21.2 | |
| ProPosNumber of clusters=25K, Beta factor=12023.03 | 52.5 | — | — | — | |
| TEMI DINOArchitecture=Resnet50, Number of clusters=25K, Beta factor=12023.03 | 51.8 | — | — | — | |
| MAE-CTBackbone=ViT-B/16, Augmentation=minimal image augmentations, variant=MAE-CTmin2023.04 | 45.4 | 35 | 68 | 19.5 | |
| PCLNumber of clusters=25K, Beta factor=12023.03 | 41 | — | — | — | |
| MoCoNumber of clusters=25K, Beta factor=12023.03 | 28.5 | — | — | — | |
| DeepClusterNumber of clusters=25K, Beta factor=12023.03 | 28.1 | — | — | — | |
| MAE (reimpl.)Backbone=ViT-L/16, Augmentation=minimal image augmentations2023.04 | 22.9 | 13.9 | 51.2 | 4.5 | |
| MAE (reimpl.)Backbone=ViT-B/16, Augmentation=minimal image augmentations2023.04 | 20.8 | 13.5 | 51 | 4.3 | |
| MAE (reimpl.)Backbone=ViT-H/16, Augmentation=minimal image augmentations2023.04 | 17 | 10.6 | 47.5 | 2.7 | |
| SCANArch.=Resnet502023.03 | — | 39.9 | — | 27.5 | |
| SeLaArch.=Resnet502023.03 | — | 30.5 | — | 16.2 | |
| SSCNArch.=Resnet502023.03 | — | 41.1 | — | 29.5 | |
| TEMI DINOArch.=Resnet502023.03 | — | 45.2 | — | 31.3 | |
| TEMI DINOArch.=ViT-B/162023.03 | — | 58.4 | — | 45.9 | |
| TEMI MSNArch.=ViT-L/162023.03 | — | 61.6 | — | 48.4 |