Image Classification on ImageNet-C (val)
33.8mCEMAE
Evaluation Results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MAEModel=ViT-H, Resolution=224, Training Objective=MAE pre-training2021.11 | 33.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAEModel=ViT-H, Resolution=448, Training Objective=MAE pre-training2021.11 | 36.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Source only2025.09 | 38.7 | 24.3 | 26.1 | 25.1 | 22.2 | 23.2 | 35.4 | 43.2 | 48.4 | 49.3 | 56.9 | 70 | 26.8 | 45.1 | 43.7 | 41.4 | — | |
| LAME2025.09 | 38.8 | 24.3 | 26.1 | 25.1 | 22.2 | 23.2 | 35.4 | 43.2 | 48.4 | 49.3 | 56.9 | 70 | 26.8 | 45.1 | 43.7 | 41.4 | — | |
| ETA2025.09 | 40.8 | 26.4 | 28.4 | 27.2 | 23.5 | 24.6 | 37.2 | 45.1 | 51 | 50.8 | 58.8 | 70.6 | 29.1 | 47.7 | 46.9 | 45 | — | |
| Supervised BaselineModel=ViT-H, Resolution=224, Training Objective=Supervised2021.11 | 41.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SAR2025.09 | 41.5 | 26.5 | 29.2 | 28 | 24.5 | 25.3 | 37.4 | 45.1 | 51.7 | 51 | 59.1 | 70.5 | 31.5 | 48.2 | 48.6 | 46.3 | — | |
| SHOT2025.09 | 41.7 | 28 | 30.1 | 28.8 | 25 | 26 | 38 | 45.7 | 51.5 | 51 | 59.1 | 70.6 | 30.2 | 48.4 | 47.8 | 45.8 | — | |
| Tent2025.09 | 41.7 | 27.8 | 30 | 28.8 | 24.9 | 25.9 | 38 | 45.5 | 51.3 | 51 | 59.1 | 70.6 | 30 | 48.2 | 47.8 | 45.7 | — | |
| MAEModel=ViT-L, Resolution=224, Training Objective=MAE pre-training2021.11 | 41.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Supervised BaselineModel=ViT-L, Resolution=224, Training Objective=Supervised2021.11 | 42.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepAugmentNotes=Previous Best2021.11 | 42.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| STL (FAN-L-Hybrid)Params (M)=77.32024.01 | 42.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FAN-L-HybridParams (M)=76.82024.01 | 43 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransNeXt-BaseResolution=224x224, Parameters=89.7M, FLOPs=18.4G2023.11 | 43.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| STL (FAN-B-Hybrid)Params (M)=50.92024.01 | 43.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MaxViT-BaseResolution=224x224, Parameters=119.5M, FLOPs=24.0G2023.11 | 43.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransNeXt-SmallResolution=224x224, Parameters=49.7M, FLOPs=10.3G2023.11 | 43.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DAT-AugReg-ViTParams (M)=862024.01 | 44.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FAN-B-HybridParams (M)=50.42024.01 | 45.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Supervised BaselineModel=ViT-B, Resolution=224, Training Objective=Supervised2021.11 | 45.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PVTv2-B5Resolution=224x224, Parameters=82.0M, FLOPs=11.8G2023.11 | 45.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MaxViT-SmallResolution=224x224, Parameters=68.9M, FLOPs=11.7G2023.11 | 46.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransNeXt-TinyResolution=224x224, Parameters=28.2M, FLOPs=5.7G2023.11 | 46.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PVTv2-B4Resolution=224x224, Parameters=62.6M, FLOPs=10.1G2023.11 | 46.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MSNArchitecture=ViT-B/16, Pre-training method=MSN2022.04 | 46.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ConvNext-BParams (M)=88.62024.01 | 46.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RVT-BParams (M)=91.82024.01 | 46.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ConvNeXt-BResolution=224x224, Parameters=88.6M, FLOPs=15.4G2023.11 | 46.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BiFormer-BResolution=224x224, Parameters=56.8M, FLOPs=9.8G2023.11 | 47.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| STL (FAN-S-Hybrid)Params (M)=26.52024.01 | 47.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FAN-S-HybridParams (M)=26.32024.01 | 47.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PVTv2-B3Resolution=224x224, Parameters=45.2M, FLOPs=6.9G2023.11 | 48 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| iBOTBackbone=ViT-S/16, Pre-training epochs=800, Evaluation protocol=Linear Probing2021.11 | 48.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BiFormer-SResolution=224x224, Parameters=25.5M, FLOPs=4.5G2023.11 | 48.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeiT-Base/16Resolution=224x224, Parameters=86.6M, FLOPs=17.6G2023.11 | 48.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| XCIT-S24Params (M)=47.72024.01 | 49.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ConvNeXt-SResolution=224x224, Parameters=50.2M, FLOPs=8.7G2023.11 | 49.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FocalNet-B (LRF)Resolution=224x224, Parameters=88.7M, FLOPs=15.4G2023.11 | 49.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MaxViT-TinyResolution=224x224, Parameters=30.9M, FLOPs=5.6G2023.11 | 49.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LV-ViT-MParams (M)=562024.01 | 50.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TransNeXt-MicroResolution=224x224, Parameters=12.8M, FLOPs=2.7G2023.11 | 50.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FocalNet-S (LRF)Resolution=224x224, Parameters=50.3M, FLOPs=8.7G2023.11 | 51 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ConvNext-SParams (M)=50.22024.01 | 51.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RVT-SParams (M)=23.32024.01 | 51.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| XCIT-S12Params (M)=26.32024.01 | 51.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAEModel=ViT-B, Resolution=224, Training Objective=MAE pre-training2021.11 | 51.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DINOBackbone=ViT-S/16, Pre-training epochs=800, Evaluation protocol=Linear Probing2021.11 | 51.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAE-ViT-BParams (M)=862024.01 | 51.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAEArchitecture=ViT-B/16, Pre-training method=MAE2022.04 | 51.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PVTv2-B2Resolution=224x224, Parameters=25.4M, FLOPs=4.0G2023.11 | 52.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Swin-SParams (M)=502024.01 | 52.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LV-ViT-SParams (M)=262024.01 | 52.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ConvNext-TParams (M)=28.62024.01 | 53.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ConvNeXt-TResolution=224x224, Parameters=28.6M, FLOPs=4.5G2023.11 | 53.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TRUST naive2025.09 | 53.4 | 43.4 | 45.6 | 44.9 | 38.3 | 36.6 | 53 | 54.9 | 60.2 | 57.1 | 66 | 72.2 | 50.2 | 59 | 61.1 | 58.5 | — | |
| Swin-BParams (M)=87.82024.01 | 54.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Swin-BResolution=224x224, Parameters=87.8M, FLOPs=15.4G2023.11 | 54.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeiT-Small/16Resolution=224x224, Parameters=22.1M, FLOPs=4.6G2023.11 | 54.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Swin-SResolution=224x224, Parameters=49.6M, FLOPs=8.7G2023.11 | 54.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FocalNet-T (LRF)Resolution=224x224, Parameters=28.6M, FLOPs=4.5G2023.11 | 55 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CvT*-LS-21Params (M)=32.12021.07 | 55.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BiFormer-TResolution=224x224, Parameters=13.1M, FLOPs=2.2G2023.11 | 55.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TRUST2025.09 | 56.1 | 46.8 | 49.4 | 48.5 | 42.8 | 40.8 | 57.1 | 57.9 | 61.7 | 57.3 | 66.8 | 71.9 | 54.9 | 61.4 | 63.6 | 60.2 | — | |
| CvT-21Params (M)=322021.07 | 56.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeiT-SParams (M)=22.12021.07 | 57.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CvT*-LS-13Params (M)=20.32021.07 | 58.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LSNet-BFLOPS=1.3G2025.03 | 59.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CvT-13Params (M)=202021.07 | 59.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Swin-TParams (M)=28.32024.01 | 59.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Swin-TResolution=224x224, Parameters=28.3M, FLOPs=4.5G2023.11 | 62 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PVTv2-B1FLOPS=2.1G2025.03 | 62.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PVTv2-B1Resolution=224x224, Parameters=14.0M, FLOPs=2.1G2023.11 | 62.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FastViT-T12FLOPS=1.4G2025.03 | 64.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LSNet-SFLOPS=0.5G2025.03 | 65.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UniRepLKNet-AFLOPS=0.6G2025.03 | 67 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PoolFormer-S12FLOPS=1.8G2025.03 | 67.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LSNet-TFLOPS=0.3G2025.03 | 68.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FasterNet-T2FLOPS=1.9G2025.03 | 70.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientViT-M3FLOPS=0.3G2025.03 | 71.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FastViT-T8FLOPS=0.7G2025.03 | 72.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EdgeNeXt-SFLOPS=1.3G2025.03 | 72.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RN101Backbone=ResNet-1012026.01 | 73.1 | 68.3 | 69.1 | 67.8 | 74.2 | 70.5 | 71.8 | 72.3 | 73.5 | 72.1 | 75.8 | 79.2 | 76.4 | 73.9 | 74.6 | 77.1 | — | |
| PVTv2-B0FLOPS=0.6G2025.03 | 75.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ViTBackbone=ViT2026.01 | 75.7 | 71.2 | 72 | 70.5 | 76.8 | 73.4 | 74.5 | 75.1 | 76.2 | 74.8 | 78.3 | 81.5 | 78.9 | 76.5 | 77.3 | 79.6 | — | |
| ResNet50Architecture=ResNet50, Pre-training method=Supervised2022.04 | 76.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| StarNet-S1FLOPS=0.4G2025.03 | 77.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SwinBackbone=Swin2026.01 | 78.4 | 74.5 | 75.2 | 73.8 | 79.1 | 76.2 | 77.3 | 78 | 78.9 | 77.5 | 80.6 | 83.2 | 81.3 | 79.2 | 79.8 | 82 | — | |
| ResNet-50Params (M)=25.62021.07 | 78.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PVT-TinyResolution=224x224, Parameters=13.2M, FLOPs=1.9G2023.11 | 79.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PVT-TinyFLOPS=1.9G2025.03 | 79.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HVTBackbone=HVT2026.01 | 81.6 | 78.9 | 79.3 | 77.6 | 82.4 | 79.8 | 80.7 | 81.2 | 81.8 | 80.4 | 83.1 | 85.6 | 84 | 82.1 | 82.5 | 84.8 | — | |
| MoCo-v2 + Patch-base NSalpha=2, k=327682021.10 | 87.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MoCo-v2 + Patch-base NSalpha=2, k=163842021.10 | 87.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Patch-base NSalpha=2, base_method=MoCo-v22021.10 | 87.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MoCo-v2 + Patch-base NSalpha=2, k=655362021.10 | 87.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MoCo-v22021.10 | 87.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MoCo-v22021.10 | 87.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EdgeNeXt-XSFLOPS=0.5G2025.03 | 88.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MoCHibase_method=MoCo-v22021.10 | 88.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |