Image Classification on ImageNet 1k (val) (k-NN and Linear Probing)
80.5Linear Probing AccuracyiBOT-vMF + KP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| iBOT-vMF + KPBackbone=ViT-B/16, Epochs=4002025.10 | 80.5 | 78.8 | |
| iBOT-vMFBackbone=ViT-B/16, Epochs=4002025.10 | 80.3 | 78.7 | |
| iBOTBackbone=ViT-B/16, Epochs=4002025.10 | 79.5 | 77.1 | |
| DINO-vMFBackbone=ViT-B/16, Epochs=4002025.10 | 78.8 | 77.4 | |
| DINOBackbone=ViT-B/16, Epochs=4002025.10 | 78.2 | 76.1 | |
| CARP + DecouplingBackbone=ViT-B/16, Epochs=4002025.10 | 78.1 | 76.7 | |
| iBOTBackbone=ViT-S/16, Epochs=8002025.10 | 77.9 | 75.2 | |
| iBOT-vMFBackbone=ViT-S/16, Epochs=8002025.10 | 77.9 | 75.3 | |
| iBOT-vMF + KPBackbone=ViT-S/16, Epochs=8002025.10 | 77.9 | 75.3 | |
| iBOTBackbone=ViT-S/16, Epochs=3002025.10 | 77.4 | 74.6 | |
| DINOBackbone=ViT-S/16, Epochs=8002025.10 | 77 | 74.5 | |
| DINO-vMFBackbone=ViT-S/16, Epochs=8002025.10 | 77 | 74.7 | |
| MSNBackbone=ViT-S/16, Epochs=6002025.10 | 76.9 | — | |
| CARP + DecouplingBackbone=ViT-S/16, Epochs=8002025.10 | 76.4 | 75.3 | |
| CARPBackbone=ViT-S/16, Epochs=3002025.10 | 76.3 | 73.6 | |
| DINOBackbone=ViT-S/16, Epochs=3002025.10 | 76.2 | 72.8 | |
| CARP + DecouplingBackbone=ViT-S/16, Epochs=3002025.10 | 76.2 | 74.1 | |
| CARP + KPBackbone=ViT-S/16, Epochs=3002025.10 | 76.1 | 73.7 | |
| DINOBackbone=RN-50, Epochs=4002025.10 | 75.3 | 67.5 | |
| CARPBackbone=RN-50, Epochs=4002025.10 | 75.3 | 67.7 | |
| CARP + DecouplingBackbone=RN-50, Epochs=4002025.10 | 75.3 | 69.1 | |
| SWAVBackbone=RN-50, Epochs=4002025.10 | 74.6 | 65 |