Image Classification on CIFAR-100 (Label Noise Robustness)
70.1Test Accuracy (Clean)DM
Evaluation results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DMBackbone=ResNet-44, lambda=02019.05 | 70.1 | 60.9 | 55.2 | 44.6 | 65.5 | 63.1 | 60.6 | — | — | — | — | — | — | — | — | — | |
| CCEBackbone=ResNet-44, Batch size=256, Optimizer=SGD, Weight decay=1e-4, Training iterations=30k2019.05 | 70 | 60.4 | 53.2 | 42.1 | 66.4 | 64.7 | 60.3 | — | — | — | — | — | — | — | — | — | |
| DMBackbone=ResNet-44, lambda=0.52019.05 | 69.3 | 65.7 | 61 | 52.9 | 67.4 | 65 | 60.8 | — | — | — | — | — | — | — | — | — | |
| DMBackbone=ResNet-44, lambda=12019.05 | 69.2 | 63.4 | 54.7 | 43.9 | 67.5 | 65.8 | 63.3 | — | — | — | — | — | — | — | — | — | |
| CCE-DNBackbone=ResNet-44, DN variant=true2019.05 | 69.1 | 60.7 | 54.2 | 44.6 | 65.9 | 64 | 60.5 | — | — | — | — | — | — | — | — | — | |
| DMBackbone=ResNet-44, beta=02019.05 | 67.2 | 56.2 | 50.9 | 44.4 | 64.4 | 62.5 | 60.4 | — | — | — | — | — | — | — | — | — | |
| SLBackbone=ResNet-44, Source=Wang et al., 2019c2019.05 | 66.8 | 60 | 53.7 | 41.5 | 65.6 | 65.1 | 63.1 | — | — | — | — | — | — | — | — | — | |
| GCE-DNBackbone=ResNet-44, DN variant=true2019.05 | 65.8 | 62.5 | 58.3 | 48.4 | 64.1 | 62.1 | 60.3 | — | — | — | — | — | — | — | — | — | |
| D2LBackbone=ResNet-44, Batch size=256, Optimizer=SGD, Source=Wang et al., 2019c2019.05 | 64.6 | 59.2 | 52 | 35.3 | 62.4 | 63.2 | 61.4 | — | — | — | — | — | — | — | — | — | |
| ForwardBackbone=ResNet-44, Batch size=256, Optimizer=SGD, Source=Wang et al., 2019c2019.05 | 64 | 59.8 | 53.1 | 24.7 | 64.1 | 64 | 60.9 | — | — | — | — | — | — | — | — | — | |
| LSBackbone=ResNet-44, Batch size=256, Optimizer=SGD, Weight decay=1e-4, Training iterations=30k, Source=Wang et al., 2019c2019.05 | 63.7 | 58.8 | 50.1 | 24.7 | 63 | 62.3 | 61.6 | — | — | — | — | — | — | — | — | — | |
| GCEBackbone=ResNet-44, Batch size=256, Optimizer=SGD2019.05 | 63.6 | 62.4 | 58.6 | 50.6 | 62.8 | 62.2 | 58.7 | — | — | — | — | — | — | — | — | — | |
| Boot-hardBackbone=ResNet-44, Batch size=256, Optimizer=SGD, Weight decay=1e-4, Source=Wang et al., 2019c2019.05 | 63.3 | 57.9 | 48.2 | 12.3 | 63.4 | 63.2 | 62.1 | — | — | — | — | — | — | — | — | — | |
| MSEBackbone=ResNet-442019.05 | 28 | 24.6 | 21.3 | 18 | 24.5 | 24.3 | 23 | — | — | — | — | — | — | — | — | — | |
| MSE-DNBackbone=ResNet-44, DN variant=true2019.05 | 25.8 | 28.4 | 27 | 26.5 | 27.4 | 23.9 | 25.3 | — | — | — | — | — | — | — | — | — | |
| MAEBackbone=ResNet-44, Batch size=2562019.05 | 8.2 | 6.4 | 7.3 | 5.2 | 7.3 | 6.3 | 7.3 | — | — | — | — | — | — | — | — | — | |
| MAE-DNBackbone=ResNet-44, DN variant=true2019.05 | 7.5 | 5.4 | 4.5 | 4.8 | 5.8 | 3.5 | 3.9 | — | — | — | — | — | — | — | — | — | |
| AoSZero-shot=true, Adversarial fine-tuning=ImageNet, PGD attack steps=20, Perturbation radius (epsilon)=1/2552026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 40.23 | |
| AUMBackbone=ViT-Small2026.02 | — | — | — | — | — | — | — | — | — | — | 22.58 | 15.32 | 12.88 | 10.15 | 6.08 | — | |
| BSAMBackbone=ResNet-18, Noise Type=Symmetric2026.01 | — | — | — | — | — | — | — | 71.65 | 57.61 | 38.71 | 32.69 | — | — | — | — | — | |
| CCSAUMBackbone=ViT-Small2026.02 | — | — | — | — | — | — | — | — | — | — | 59.18 | 55.35 | 51.16 | 27.13 | 22.84 | — | |
| CLIPZero-shot=true, Adversarial fine-tuning=ImageNet, PGD attack steps=20, Perturbation radius (epsilon)=1/2552026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 4.85 | |
| DCQBackbone=ViT-Small2026.02 | — | — | — | — | — | — | — | — | — | — | 68.38 | 65.19 | 63.72 | 47.19 | 35.42 | — | |
| EL2NBackbone=ViT-Small2026.02 | — | — | — | — | — | — | — | — | — | — | 22.49 | 14.55 | 11.67 | 9.18 | 8.76 | — | |
| FAREZero-shot=true, Adversarial fine-tuning=ImageNet, PGD attack steps=20, Perturbation radius (epsilon)=1/2552026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 38.98 | |
| NCSAMBackbone=ResNet-18, Noise Type=Symmetric2026.01 | — | — | — | — | — | — | — | 76.05 | 70.59 | 58.01 | 39.02 | — | — | — | — | — | |
| OursZero-shot=true, Adversarial fine-tuning=ImageNet, PGD attack steps=20, Perturbation radius (epsilon)=1/2552026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 41.89 | |
| Ours (5 trees)Zero-shot=true, Adversarial fine-tuning=ImageNet, PGD attack steps=20, Perturbation radius (epsilon)=1/255, Number of trees=52026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 42.66 | |
| PMG-FTZero-shot=true, Adversarial fine-tuning=ImageNet, PGD attack steps=20, Perturbation radius (epsilon)=1/2552026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 35.92 | |
| RandomBackbone=ViT-Small2026.02 | — | — | — | — | — | — | — | — | — | — | 33.83 | 30.96 | 26.19 | 25.61 | 22.68 | — | |
| SAMBackbone=ResNet-18, Noise Type=Symmetric2026.01 | — | — | — | — | — | — | — | 69.1 | 55.06 | 35.59 | 9.82 | — | — | — | — | — | |
| SGDBackbone=ResNet-18, Noise Type=Symmetric2026.01 | — | — | — | — | — | — | — | 65.42 | 48.91 | 31.46 | 12.32 | — | — | — | — | — | |
| TeCoAZero-shot=true, Adversarial fine-tuning=ImageNet, PGD attack steps=20, Perturbation radius (epsilon)=1/2552026.04 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 34.16 |