Image Classification on ImageNet 1.0 (10% labeled)
85.9AccuracyDHO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DHOArchitecture=ViT-L/14, Params (M)=3042025.05 | 85.9 | — | — | |
| MCTArchitecture=Multi-arch (4), Params (M)=10712025.05 | 85.8 | — | — | |
| MCTArchitecture=Multi-arch (2), Params (M)=6082025.05 | 85.2 | — | — | |
| REACTArchitecture=ViT-L/14, Params (M)=3042025.05 | 85.1 | — | — | |
| CTArchitecture=Multi-arch (2), Params (M)=6082025.05 | 85.1 | — | — | |
| REACT (Locked-Text)Backbone=ViT-L/14, # Params=304M, Initialization=Locked-Text2023.01 | 85.1 | — | — | |
| REACT (Gated-Image)Architecture=ViT-L/14, Params (M)=3802025.05 | 85 | — | — | |
| REACT (Locked-Text Gated-Image)Backbone=ViT-L/14, # Params=380M, Initialization=Locked-Text Gated-Image2023.01 | 85 | — | — | |
| CTArchitecture=Multi-arch (4), Params (M)=10712025.05 | 84.8 | — | — | |
| CLIPArchitecture=ViT-L/14, Params (M)=3042025.05 | 84.7 | — | — | |
| CLIP (Language Init.)Backbone=ViT-L/14, # Params=304M, Initialization=Language Init.2023.01 | 84.7 | — | — | |
| Semi-ViTArchitecture=ViT-H/14, Params (M)=6322025.05 | 84.3 | — | — | |
| Semi-ViTBackbone=ViT-H/14, # Params=632M2023.01 | 84.3 | — | — | |
| CE + KD (logit)Architecture=ViT-L/14, Params (M)=3042025.05 | 83.6 | — | — | |
| Semi-ViTArchitecture=ViT-L/14, Params (M)=3072025.05 | 83.3 | — | — | |
| DHOArchitecture=ViT-B/16, Params (M)=862025.05 | 82.8 | — | — | |
| REACT (Gated-Image)Architecture=ViT-B/16, Params (M)=1292025.05 | 81.8 | — | — | |
| REACT (Locked-Text Gated-Image)Backbone=ViT-B/16, # Params=129M, Initialization=Locked-Text Gated-Image2023.01 | 81.8 | — | — | |
| SimCLR v2 self-distillArchitecture=ResNet-154 (3x+SK), Params (M)=7952025.05 | 80.9 | — | — | |
| SimCLRv2 (self-distilled)Backbone=RN152x3, # Params=795M2023.01 | 80.9 | — | — | |
| REACTArchitecture=ViT-B/16, Params (M)=862025.05 | 80.8 | — | — | |
| REACT (Locked-Text)Backbone=ViT-B/16, # Params=86M, Initialization=Locked-Text2023.01 | 80.8 | — | — | |
| CLIPArchitecture=ViT-B/16, Params (M)=862025.05 | 80.4 | — | — | |
| CE + KD (logit)Architecture=ViT-B/16, Params (M)=862025.05 | 80.4 | — | — | |
| CLIP (Language Init.)Backbone=ViT-B/16, # Params=86M, Initialization=Language Init.2023.01 | 80.4 | — | — | |
| MSNArchitecture=ViT-B/4, Params (M)=862025.05 | 80.2 | — | — | |
| SimCLR v2 distillArchitecture=ResNet-50 (2x+SK), Params (M)=1402025.05 | 80.2 | — | — | |
| SimCLRv2Backbone=RN152x3, # Params=795M2023.01 | 80.1 | — | — | |
| CLIP (Random Init.)Backbone=ViT-B/16, # Params=86M, Initialization=Random Init.2023.01 | 80.1 | — | — | |
| PAWSBackbone=RN50x4, # Params=375M2023.01 | 79 | — | — | |
| SEERBackbone=RegNetY, # Params=1.3B2023.01 | 77.9 | — | — | |
| CLIP (Zero-Shot)Backbone=ViT-L/14, # Params=304M, Protocol=Zero-Shot2023.01 | 75.3 | — | — | |
| CLIP (Zero-Shot)Backbone=ViT-B/16, # Params=86M, Protocol=Zero-Shot2023.01 | 68.6 | — | — | |
| Barlow TwinsBackbone=resnet-50 x12022.07 | — | 69.7 | 89.3 | |
| BYOLBackbone=resnet-50 x12022.07 | — | 68.8 | 89 | |
| BYOLBackbone=resnet-50 x22022.07 | — | 73.5 | 91.7 | |
| NNCLRBackbone=resnet-50 x12022.07 | — | 69.8 | 89.3 | |
| SimCLRBackbone=resnet-50 x12022.07 | — | 65.6 | 87.8 | |
| SimCLRBackbone=resnet-50 x22022.07 | — | 71.7 | 91.2 | |
| SMoGBackbone=resnet-50 x12022.07 | — | 71.2 | 90.5 | |
| SMoGBackbone=resnet-50 x22022.07 | — | 74.4 | 92.4 | |
| SSL-HSICBackbone=resnet-50 x12022.07 | — | 67.9 | 88.6 | |
| SupervisedBackbone=resnet-50 x12022.07 | — | 56.4 | 80.4 | |
| SWAVBackbone=resnet-50 x12022.07 | — | 70.2 | 89.9 |