Image Classification on ImageNet 1K (val) (FT, Linear, and Attentive Probing Accuracy)
86.3FT Top-1 AccCAE*
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CAE*Backbone=ViT-L, #Epochs=1600, #Crops=1, Tokenizer=DALL-E2022.02 | 86.3 | 78.1 | 81.2 | |
| CAEBackbone=ViT-L, #Epochs=1600, #Crops=1, Tokenizer=d-VAE2022.02 | 86.3 | 77.9 | 81.2 | |
| MAEBackbone=ViT-L, #Epochs=1600, #Crops=12022.02 | 86 | 76 | 78.8 | |
| BEITBackbone=ViT-L, #Epochs=1600, #Crops=12022.02 | 85.2 | — | — | |
| MoCo v3Backbone=ViT-L, #Epochs=600+, #Crops=22022.02 | 84.1 | — | — | |
| CAE*Backbone=ViT-B, #Epochs=1600, #Crops=1, Tokenizer=DALL-E2022.02 | 83.9 | 70.4 | 77.1 | |
| CAEBackbone=ViT-B, #Epochs=1600, #Crops=1, Tokenizer=d-VAE2022.02 | 83.9 | 71.4 | 77.4 | |
| SimMIMBackbone=ViT-B, #Epochs=800, #Crops=12022.02 | 83.8 | 56.7 | — | |
| iBOTBackbone=ViT-B, #Epochs=1600+, #Crops=122022.02 | 83.8 | 79.5 | 79.8 | |
| CAE*Backbone=ViT-B, #Epochs=800, #Crops=1, Tokenizer=DALL-E2022.02 | 83.8 | 68.6 | 75.9 | |
| MAEBackbone=ViT-B, #Epochs=1600, #Crops=12022.02 | 83.6 | 67.8 | 74.2 | |
| CAE*Backbone=ViT-B, #Epochs=300, #Crops=1, Tokenizer=DALL-E2022.02 | 83.6 | 64.1 | 73.8 | |
| DINOBackbone=ViT-B, #Epochs=1600+, #Crops=122022.02 | 83.3 | 77.3 | 77.8 | |
| MoCo v3Backbone=ViT-B, #Epochs=600+, #Crops=22022.02 | 83 | 76.2 | 77 | |
| BEITBackbone=ViT-B, #Epochs=300, #Crops=12022.02 | 83 | 37.6 | 49.4 | |
| MAEBackbone=ViT-B, #Epochs=300, #Crops=12022.02 | 82.9 | 61.5 | 71.1 | |
| CAE*Backbone=ViT-S, #Epochs=300, #Crops=1, Tokenizer=DALL-E2022.02 | 82 | 51.8 | 65 | |
| MoCo v3Backbone=ViT-S, #Epochs=600+, #Crops=22022.02 | 81.7 | 73.1 | 73.8 | |
| BEITBackbone=ViT-S, #Epochs=300, #Crops=12022.02 | 81.7 | 15.7 | 23.6 | |
| DeiTBackbone=ViT-S, #Epochs=3002022.02 | — | — | 79.9 | |
| DeiTBackbone=ViT-B, #Epochs=3002022.02 | — | — | 81.8 |