Image Classification on ImageNet-1k (Top-1 Accuracy & GMACs)
87.31Top-1 AccuracyDINOv3
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DINOv3Size=L, Evaluation Protocol=Linear Probing2026.07 | 87.31 | — | |
| DINOv2Size=L, Evaluation Protocol=Linear Probing2026.07 | 86.43 | — | |
| LingBot-VisionSize=L, Evaluation Protocol=Linear Probing2026.07 | 86.38 | — | |
| SigLIP2Size=L, Evaluation Protocol=Linear Probing2026.07 | 85.67 | — | |
| DINOv3Size=L, Evaluation Protocol=k-NN2026.07 | 85.27 | — | |
| LingBot-VisionSize=B, Evaluation Protocol=Linear Probing2026.07 | 85.05 | — | |
| DINOv3Size=B, Evaluation Protocol=Linear Probing2026.07 | 84.79 | — | |
| DINOv2Size=B, Evaluation Protocol=Linear Probing2026.07 | 84.27 | — | |
| DINOv2Size=L, Evaluation Protocol=k-NN2026.07 | 83.82 | — | |
| LingBot-VisionSize=L, Evaluation Protocol=k-NN2026.07 | 83.62 | — | |
| DINOv3Size=B, Evaluation Protocol=k-NN2026.07 | 83.21 | — | |
| SigLIP2Size=L, Evaluation Protocol=k-NN2026.07 | 82.91 | — | |
| Slimmable ConvNeXt-BBackbone=ConvNeXt-B, Number of subnetworks=4, Slimming ratio (p)=0.75, Training epochs=6002026.05 | 82.8 | 8.8 | |
| Slimmable ConvNeXt-BBackbone=ConvNeXt-B, Number of subnetworks=4, Slimming ratio (p)=1.0, Training epochs=6002026.05 | 82.8 | 15.4 | |
| Slimmable ConvNeXt-BBackbone=ConvNeXt-B, Number of subnetworks=3, Slimming ratio (p)=1.0, Training epochs=6002026.05 | 82.5 | 15.4 | |
| FrancaSize=L, Evaluation Protocol=Linear Probing2026.07 | 82.5 | — | |
| SigLIP2Size=B, Evaluation Protocol=Linear Probing2026.07 | 82.47 | — | |
| LingBot-VisionSize=B, Evaluation Protocol=k-NN2026.07 | 82.32 | — | |
| Slimmable ConvNeXt-SBackbone=ConvNeXt-S, Number of subnetworks=3, Slimming ratio (p)=1.0, Training epochs=6002026.05 | 82.3 | 8.7 | |
| LingBot-VisionSize=S, Evaluation Protocol=Linear Probing2026.07 | 82.22 | — | |
| Slimmable ConvNeXt-SBackbone=ConvNeXt-S, Number of subnetworks=4, Slimming ratio (p)=1.0, Training epochs=6002026.05 | 82.2 | 8.7 | |
| DINOv2Size=B, Evaluation Protocol=k-NN2026.07 | 82.17 | — | |
| MatFormerBackbone=ViT-B, Slimming ratio (p)=0.75, Training epochs=6002026.05 | 82 | 14.8 | |
| MatFormerBackbone=ViT-B, Slimming ratio (p)=1.0, Training epochs=6002026.05 | 82 | 17.6 | |
| Slimmable ConvNeXt-SBackbone=ConvNeXt-S, Number of subnetworks=4, Slimming ratio (p)=0.75, Training epochs=6002026.05 | 81.9 | 5 | |
| Slimmable ConvNeXt-BBackbone=ConvNeXt-B, Number of subnetworks=3, Slimming ratio (p)=0.5, Training epochs=6002026.05 | 81.8 | 4 | |
| MatFormerBackbone=ViT-B, Slimming ratio (p)=0.5, Training epochs=6002026.05 | 81.8 | 12 | |
| Slimmable ConvNeXt-BBackbone=ConvNeXt-B, Number of subnetworks=4, Slimming ratio (p)=0.5, Training epochs=6002026.05 | 81.6 | 4 | |
| HydraViTBackbone=ViT-B, Slimming ratio (p)=1.0, Training epochs=11002026.05 | 81.6 | 17.6 | |
| HydraViTBackbone=ViT-B, Slimming ratio (p)=0.75, Training epochs=11002026.05 | 81.5 | 10 | |
| DynaBERTBackbone=ViT-B, Slimming ratio (p)=1.0, Training epochs=6002026.05 | 81.3 | 17.6 | |
| DynaBERTBackbone=ViT-B, Slimming ratio (p)=0.75, Training epochs=6002026.05 | 81.2 | 12.2 | |
| HydraViTBackbone=ViT-B, Slimming ratio (p)=1.0, Training epochs=6002026.05 | 81.1 | 17.6 | |
| HydraViTBackbone=ViT-B, Slimming ratio (p)=0.75, Training epochs=6002026.05 | 81 | 10 | |
| SortedNetBackbone=ViT-B, Slimming ratio (p)=1.0, Training epochs=6002026.05 | 80.8 | 17.6 | |
| DINOv2Size=S, Evaluation Protocol=Linear Probing2026.07 | 80.76 | — | |
| SortedNetBackbone=ViT-B, Slimming ratio (p)=0.75, Training epochs=6002026.05 | 80.6 | 10 | |
| MatFormerBackbone=ViT-B, Slimming ratio (p)=0.25, Training epochs=6002026.05 | 80.5 | 9.2 | |
| DINOv3Size=S, Evaluation Protocol=Linear Probing2026.07 | 80.37 | — | |
| Slimmable ConvNeXt-SBackbone=ConvNeXt-S, Number of subnetworks=3, Slimming ratio (p)=0.5, Training epochs=6002026.05 | 80.2 | 2.3 | |
| DynaBERTBackbone=ViT-B, Slimming ratio (p)=0.5, Training epochs=6002026.05 | 80.2 | 7.5 | |
| HydraViTBackbone=ViT-B, Slimming ratio (p)=0.5, Training epochs=11002026.05 | 80.2 | 4.6 | |
| Slimmable ConvNeXt-SBackbone=ConvNeXt-S, Number of subnetworks=4, Slimming ratio (p)=0.5, Training epochs=6002026.05 | 79.9 | 2.3 | |
| SigLIP2Size=B, Evaluation Protocol=k-NN2026.07 | 79.47 | — | |
| DINOv3Size=S, Evaluation Protocol=k-NN2026.07 | 79.33 | — | |
| HydraViTBackbone=ViT-B, Slimming ratio (p)=0.5, Training epochs=6002026.05 | 79.3 | 4.6 | |
| DINOv2Size=S, Evaluation Protocol=k-NN2026.07 | 79.11 | — | |
| LingBot-VisionSize=S, Evaluation Protocol=k-NN2026.07 | 78.98 | — | |
| SortedNetBackbone=ViT-B, Slimming ratio (p)=0.5, Training epochs=6002026.05 | 78.9 | 4.6 | |
| FrancaSize=B, Evaluation Protocol=Linear Probing2026.07 | 78.45 | — | |
| Slimmable ConvNeXt-BBackbone=ConvNeXt-B, Number of subnetworks=3, Slimming ratio (p)=0.25, Training epochs=6002026.05 | 75.9 | 1.1 | |
| Slimmable ConvNeXt-BBackbone=ConvNeXt-B, Number of subnetworks=4, Slimming ratio (p)=0.25, Training epochs=6002026.05 | 75.4 | 1.1 | |
| DynaBERTBackbone=ViT-B, Slimming ratio (p)=0.25, Training epochs=6002026.05 | 73 | 3.4 | |
| Slimmable ConvNeXt-SBackbone=ConvNeXt-S, Number of subnetworks=3, Slimming ratio (p)=0.25, Training epochs=6002026.05 | 71.7 | 0.6 | |
| HydraViTBackbone=ViT-B, Slimming ratio (p)=0.25, Training epochs=11002026.05 | 71.7 | 1.3 | |
| FrancaSize=L, Evaluation Protocol=k-NN2026.07 | 71.7 | — | |
| HydraViTBackbone=ViT-B, Slimming ratio (p)=0.25, Training epochs=6002026.05 | 70.6 | 1.3 | |
| Slimmable ConvNeXt-SBackbone=ConvNeXt-S, Number of subnetworks=4, Slimming ratio (p)=0.25, Training epochs=6002026.05 | 70.5 | 0.6 | |
| SortedNetBackbone=ViT-B, Slimming ratio (p)=0.25, Training epochs=6002026.05 | 70.2 | 1.3 | |
| FrancaSize=B, Evaluation Protocol=k-NN2026.07 | 63.45 | — |