Image Classification on ImageNet (Top-1 and Top-5 Accuracy)
88.6Top-1 AccViT-Huge
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ViT-HugeParam=632M, Data=JFT+ImageNet2022.08 | 88.6 | — | |
| DSeq-JEPAArch.=ViT-H/16448, Epochs=300, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 87.8 | — | |
| DSeq-C-JEPAArch.=ViT-L/16, Epochs=600, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 87.2 | — | |
| I-JEPAArch.=ViT-H/16448, Epochs=300, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 87.1 | — | |
| MAE-ViT-HugeParam=632M, Data=ImageNet2022.08 | 86.9 | — | |
| C-JEPAArch.=ViT-L/16, Epochs=600, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 86.9 | — | |
| DSeq-JEPAArch.=ViT-L/16, Epochs=600, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 86.8 | — | |
| I-JEPAArch.=ViT-L/16, Epochs=600, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 86.6 | — | |
| MAEArch.=ViT-L/16, Epochs=1600, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 85.9 | — | |
| EfficientNet-L2Param=480M, Data=ImageNet2022.08 | 85.5 | 97.5 | |
| NEPAArch.=ViT-L/14, Epochs=800, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 85.3 | — | |
| ConvNeXt-LParam=198M, Data=ImageNet2022.08 | 84.3 | — | |
| Semi-ViT-HugeParam=632M, Data=10%ImageNet2022.08 | 84.3 | 96.6 | |
| DSeq-C-JEPAArch.=ViT-B/16, Epochs=600, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 84.3 | — | |
| C-JEPAArch.=ViT-B/16, Epochs=600, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 84.2 | — | |
| DSeq-JEPAArch.=ViT-B/16, Epochs=600, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 84 | — | |
| iBOTArch.=ViT-B/16, Epochs=1600, Evaluation Protocol=Fine-tune, View Data Augmentation=true2025.11 | 84 | — | |
| NEPAArch.=ViT-B/14, Epochs=1600, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 83.8 | — | |
| MAEArch.=ViT-B/16, Epochs=1600, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 83.6 | — | |
| I-JEPAArch.=ViT-B/16, Epochs=600, Evaluation Protocol=Fine-tune, View Data Augmentation=false2025.11 | 83.5 | — | |
| Swin-BParam=88M, Data=ImageNet2022.08 | 83.3 | — | |
| DINOArch.=ViT-B/16, Epochs=1600, Evaluation Protocol=Fine-tune, View Data Augmentation=true2025.11 | 82.8 | — | |
| DSeq-JEPAArch.=ViT-H/16448, Epochs=300, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 82.4 | — | |
| DeiT-BParam=86M, Data=ImageNet2022.08 | 81.8 | — | |
| SENet-154Param=146M, Data=ImageNet2022.08 | 81.3 | 95.5 | |
| I-JEPAArch.=ViT-H/16448, Epochs=300, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 81.1 | — | |
| ResNeXt-101Param=84M, Data=ImageNet2022.08 | 80.9 | 95.6 | |
| Inception-v4Param=48M, Data=ImageNet2022.08 | 80 | 95 | |
| Semi-ViT-HugeParam=632M, Data=1%ImageNet2022.08 | 80 | 93.1 | |
| iBOTArch.=ViT-B/16, Epochs=1600, Evaluation Protocol=Linear, View Data Augmentation=true2025.11 | 79.5 | — | |
| LeJEPAArch.=ViT-H/14, Epochs=100, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 79 | — | |
| Inception-v3Param=24M, Data=ImageNet2022.08 | 78.8 | 94.4 | |
| CutMixBackbone=ResNet-502022.07 | 78.6 | 94.08 | |
| DSeq-C-JEPAArch.=ViT-L/16, Epochs=600, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 78.4 | — | |
| DINOArch.=ViT-B/16, Epochs=1600, Evaluation Protocol=Linear, View Data Augmentation=true2025.11 | 78.2 | — | |
| C-JEPAArch.=ViT-L/16, Epochs=600, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 78 | — | |
| DenseNet-264Param=34M, Data=ImageNet2022.08 | 77.9 | 93.9 | |
| DSeq-JEPAArch.=ViT-L/16, Epochs=600, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 77.9 | — | |
| ResNet-152Param=60M, Data=ImageNet2022.08 | 77.8 | 93.8 | |
| LoRot-EBackbone=ResNet-502022.07 | 77.72 | 93.65 | |
| LoRot-IBackbone=ResNet-502022.07 | 77.71 | 93.6 | |
| MAEArch.=ViT-H/14, Epochs=1600, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 77.2 | — | |
| I-JEPAArch.=ViT-L/16, Epochs=600, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 77.1 | — | |
| CutoutBackbone=ResNet-502022.07 | 77.07 | 93.34 | |
| + Rot(SS)Backbone=ResNet-502022.07 | 76.79 | 93.16 | |
| + Rot(MT)Backbone=ResNet-502022.07 | 76.68 | 93.1 | |
| + Rot(DA)Backbone=ResNet-502022.07 | 76.42 | 93.06 | |
| BaselineBackbone=ResNet-502022.07 | 76.32 | 92.95 | |
| SLA+SDBackbone=ResNet-502022.07 | 76.17 | — | |
| β-DARTSParams (M)=5.52026.02 | 76.1 | 93 | |
| ResNet-50Param=26M, Data=ImageNet2022.08 | 76 | 93 | |
| MAEArch.=ViT-L/16, Epochs=1600, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 76 | — | |
| DOTSParams (M)=5.22026.02 | 75.7 | 92.6 | |
| DARTS+PTParams (M)=4.62026.02 | 75.5 | 92 | |
| MIDASParams (M)=5.22026.02 | 75.4 | 92.5 | |
| P-DARTSParams (M)=5.12026.02 | 75.3 | 92.5 | |
| SDARTS-ADVParams (M)=5.42026.02 | 74.8 | 92.2 | |
| NASNet-AParams (M)=5.32026.02 | 74 | 91.6 | |
| DSeq-C-JEPAArch.=ViT-B/16, Epochs=600, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 73.8 | — | |
| DSeq-JEPAArch.=ViT-B/16, Epochs=600, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 73.5 | — | |
| C-JEPAArch.=ViT-B/16, Epochs=600, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 73.5 | — | |
| DARTSParams (M)=4.72026.02 | 73.3 | 91.3 | |
| SNASParams (M)=4.32026.02 | 72.7 | 90.8 | |
| I-JEPAArch.=ViT-B/16, Epochs=600, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 72.4 | — | |
| MAEArch.=ViT-B/16, Epochs=1600, Evaluation Protocol=Linear, View Data Augmentation=false2025.11 | 68 | — |