Image Classification on ImageNet Clean
22.4Clean ErrorAutoAugment
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AutoAugmentBackbone=ResNet-50, Training Epochs=90, Batch Size=512, Initial Learning Rate=0.01, Learning Rate Schedule=cosine decay2021.12 | 22.4 | — | — | |
| CutoutBackbone=ResNet-50, Training Epochs=90, Batch Size=512, Initial Learning Rate=0.01, Learning Rate Schedule=cosine decay2021.12 | 22.6 | — | — | |
| PIXMIXBackbone=ResNet-50, Training Epochs=90, Batch Size=512, Initial Learning Rate=0.01, Learning Rate Schedule=cosine decay, k=4, beta=42021.12 | 22.6 | — | — | |
| MixupBackbone=ResNet-50, Training Epochs=90, Batch Size=512, Initial Learning Rate=0.01, Learning Rate Schedule=cosine decay2021.12 | 22.7 | — | — | |
| AugMixBackbone=ResNet-50, Training Epochs=90, Batch Size=512, Initial Learning Rate=0.01, Learning Rate Schedule=cosine decay2021.12 | 22.8 | — | — | |
| CutMixBackbone=ResNet-50, Training Epochs=90, Batch Size=512, Initial Learning Rate=0.01, Learning Rate Schedule=cosine decay2021.12 | 22.9 | — | — | |
| BaselineBackbone=ResNet-50, Training Epochs=90, Batch Size=512, Initial Learning Rate=0.01, Learning Rate Schedule=cosine decay2021.12 | 23.9 | — | — | |
| SINBackbone=ResNet-50, Training Epochs=90, Batch Size=512, Initial Learning Rate=0.01, Learning Rate Schedule=cosine decay2021.12 | 25.4 | — | — | |
| MixupBackbone=ResNet-50, Bayesian treatment=None2025.05 | — | 76.1 | — | |
| OPTIMA MixupBackbone=ResNet-50, Bayesian treatment=None2025.05 | — | 76.8 | — | |
| Swin-ACMoE-Top 1Top-k routing=12025.02 | — | 75.39 | 92.56 | |
| Swin-ACMoE-Top 2Top-k routing=22025.02 | — | 76.31 | 93.14 | |
| Swin-Top 1Top-k routing=12025.02 | — | 75.22 | 92.51 | |
| Swin-Top 2Top-k routing=22025.02 | — | 76.1 | 92.99 |