Image Classification on ImageNet-A (val)
90.2AccuracyCoCa
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CoCaEvaluation Protocol=Zero-shot2023.02 | 90.2 | — | |
| LiT-22BEvaluation Protocol=Zero-shot, Backbone=ViT-22B/142023.02 | 90.1 | — | |
| LiT-e/14Evaluation Protocol=Zero-shot, Backbone=ViT-e/142023.02 | 88 | — | |
| BASICEvaluation Protocol=Zero-shot2023.02 | 85.6 | — | |
| LiT-g/14Evaluation Protocol=Zero-shot, Backbone=ViT-g/142023.02 | 81.8 | — | |
| CLIPEvaluation Protocol=Zero-shot2023.02 | 77.2 | — | |
| MAEModel=ViT-H, Resolution=448, Training Objective=MAE pre-training2021.11 | 76.7 | — | |
| ALIGNEvaluation Protocol=Zero-shot2023.02 | 75.8 | — | |
| MAEModel=ViT-H, Resolution=224, Training Objective=MAE pre-training2021.11 | 68.2 | — | |
| MAEModel=ViT-L, Resolution=224, Training Objective=MAE pre-training2021.11 | 57.1 | — | |
| STL (FAN-L-Hybrid)Params (M)=77.32024.01 | 46.1 | — | |
| PaLI-17BShot Setting=0-shot2022.09 | 44.7 | — | |
| STL (FAN-B-Hybrid)Params (M)=50.92024.01 | 42.6 | — | |
| FAN-L-HybridParams (M)=76.82024.01 | 41.8 | — | |
| PaLI-15BShot Setting=0-shot2022.09 | 41.16 | — | |
| FAN-B-HybridParams (M)=50.42024.01 | 39.6 | — | |
| GLaD (Random G)Distillation Space=Random G, Distillation Algorithm=DC, Resolution=256x2562023.05 | 39.3 | — | |
| GLaD (Pokémon G)Distillation Space=Pokémon G, Distillation Algorithm=DC, Resolution=256x2562023.05 | 39.1 | — | |
| DC (Pixel)Distillation Space=Pixel, Distillation Algorithm=DC, Resolution=256x2562023.05 | 38.3 | — | |
| GLaD (FFHQ G)Distillation Space=FFHQ G, Distillation Algorithm=DC, Resolution=256x2562023.05 | 38.3 | — | |
| STL (FAN-S-Hybrid)Params (M)=26.52024.01 | 38.2 | — | |
| PaLI-3BShot Setting=0-shot2022.09 | 37.92 | — | |
| GLaD (ImageNet G)Distillation Space=ImageNet G, Distillation Algorithm=DC, Resolution=256x2562023.05 | 37.4 | — | |
| ConvNext-BParams (M)=88.62024.01 | 36.7 | — | |
| MAEModel=ViT-B, Resolution=224, Training Objective=MAE pre-training2021.11 | 35.9 | — | |
| MAE-ViT-BParams (M)=862024.01 | 35.9 | — | |
| Noisy StudentNotes=Previous Best2021.11 | 35.8 | — | |
| Swin-SParams (M)=502024.01 | 35.8 | — | |
| Swin-BParams (M)=87.82024.01 | 35.8 | — | |
| LV-ViT-MParams (M)=562024.01 | 35.2 | — | |
| LV-ViT-SParams (M)=262024.01 | 33.9 | — | |
| FAN-S-HybridParams (M)=26.32024.01 | 33.9 | — | |
| Supervised BaselineModel=ViT-H, Resolution=224, Training Objective=Supervised2021.11 | 33.1 | — | |
| DIVERAlg.=MTT, IPC=12026.05 | 31.6 | — | |
| ConvNext-SParams (M)=50.22024.01 | 31.2 | — | |
| CvT-21Params (M)=322021.07 | 31.1 | — | |
| DIVERAlg.=DC, IPC=12026.05 | 30.5 | — | |
| DAT-AugReg-ViTParams (M)=862024.01 | 30.2 | — | |
| Supervised BaselineModel=ViT-L, Resolution=224, Training Objective=Supervised2021.11 | 29.6 | — | |
| CvT*-LS-21Params (M)=32.12021.07 | 29.3 | — | |
| GLaDAlg.=MTT, IPC=12026.05 | 29.1 | — | |
| DIVERAlg.=DM, IPC=12026.05 | 28.6 | — | |
| RVT-BParams (M)=91.82024.01 | 28.5 | — | |
| XCIT-S24Params (M)=47.72024.01 | 27.8 | — | |
| GLaDAlg.=DC, IPC=12026.05 | 27.4 | — | |
| Supervised BaselineModel=ViT-B, Resolution=224, Training Objective=Supervised2021.11 | 27.2 | — | |
| CvT*-LS-13Params (M)=20.32021.07 | 27 | — | |
| RVT-SParams (M)=23.32024.01 | 25.7 | — | |
| CvT-13Params (M)=202021.07 | 25.4 | — | |
| DDAlg.=MTT, IPC=12026.05 | 25.4 | — | |
| GLaDAlg.=DM, IPC=12026.05 | 25.3 | — | |
| XCIT-S12Params (M)=26.32024.01 | 25 | — | |
| DDAlg.=DC, IPC=12026.05 | 24.9 | — | |
| ConvNext-TParams (M)=28.62024.01 | 24.2 | — | |
| DDAlg.=DM, IPC=12026.05 | 24 | — | |
| Swin-TParams (M)=28.32024.01 | 21.6 | — | |
| DINO + SDMPEvaluation protocol=end-to-end finetuning2022.06 | 21.1 | — | |
| DINOEvaluation protocol=end-to-end finetuning2022.06 | 20 | — | |
| DeiT-SParams (M)=22.12021.07 | 19 | — | |
| MoCo v3 + SDMPEvaluation protocol=end-to-end finetuning2022.06 | 18.9 | — | |
| MoCo v3Evaluation protocol=end-to-end finetuning2022.06 | 18.1 | — | |
| iBOTBackbone=ViT-S/16, Pre-training epochs=800, Evaluation protocol=Linear Probing2021.11 | 13.8 | — | |
| DINOBackbone=ViT-S/16, Pre-training epochs=800, Evaluation protocol=Linear Probing2021.11 | 12.3 | — | |
| ResNet-50Params (M)=25.62021.07 | 6.2 | — | |
| EdgeNeXt-SFLOPS=1.3G2025.03 | — | 11.9 | |
| EdgeNeXt-XSFLOPS=0.5G2025.03 | — | 6.3 | |
| EdgeNeXt-XXSFLOPS=0.3G2025.03 | — | 3.6 | |
| EfficientViT-M3FLOPS=0.3G2025.03 | — | 5.2 | |
| FasterNet-T0FLOPS=0.3G2025.03 | — | 2.3 | |
| FasterNet-T2FLOPS=1.9G2025.03 | — | 8.7 | |
| FastViT-T12FLOPS=1.4G2025.03 | — | 14 | |
| FastViT-T8FLOPS=0.7G2025.03 | — | 6.9 | |
| LSNet-BFLOPS=1.3G2025.03 | — | 17.3 | |
| LSNet-SFLOPS=0.5G2025.03 | — | 9.6 | |
| LSNet-TFLOPS=0.3G2025.03 | — | 6.7 | |
| MAEArchitecture=ViT-B/16, Pre-training method=MAE2022.04 | — | 35.9 | |
| MSNArchitecture=ViT-B/16, Pre-training method=MSN2022.04 | — | 37.5 | |
| PoolFormer-S12FLOPS=1.8G2025.03 | — | 6.9 | |
| PVT-TinyFLOPS=1.9G2025.03 | — | 7.9 | |
| PVTv2-B0FLOPS=0.6G2025.03 | — | 4.2 | |
| PVTv2-B1FLOPS=2.1G2025.03 | — | 14.6 | |
| ResNet50Architecture=ResNet50, Pre-training method=Supervised2022.04 | — | 0.04 | |
| StarNet-S1FLOPS=0.4G2025.03 | — | 4.5 | |
| UniRepLKNet-AFLOPS=0.6G2025.03 | — | 8.4 |