| PGDF (Inception-ResNet-v2) | 75.45 | 93.11 | 81.47 | 94.03 | Sample Prior Guided Robust Model Learning to Suppress Noisy Labels | |
| CoDiM-Sup (Inception-ResNet-v2) | 76.52 | 91.96 | 80.88 | 92.48 | CoDiM: Learning with Noisy Labels via Contrastive Semi-Supervised Learning | - |
| NCR+Mixup+DA (ResNet-50) | - | - | 80.5 | - | Learning with Neighbor Consistency for Noisy Labels | |
| Dynamic Loss (Inception-ResNet-v2) | 74.76 | 93.08 | 80.12 | 93.64 | Dynamic Loss For Robust Learning | |
| CoDiM-Self (Inception-ResNet-v2) | 77.24 | 92.48 | 80.12 | 93.52 | CoDiM: Learning with Noisy Labels via Contrastive Semi-Supervised Learning | - |
| Sel-CL+ (ResNet-18) | 76.84 | 93.04 | 79.96 | 92.64 | Selective-Supervised Contrastive Learning with Noisy Labels | |
| CPC | 75.75±0.14 | 93.49±0.25 | 79.63±0.08 | 93.46±0.10 | Class Prototype-based Cleaner for Label Noise Learning | |
| DivideMix with C2D (ResNet-50) | 78.57 ± 0.37 | 93.04 ± 0.10 | 79.42 ± 0.34 | 92.32 ± 0.33 | Contrast to Divide: Self-Supervised Pre-Training for Learning with Noisy Labels | |
| NGC (Inception-ResNet-v2) | 74.44 | 91.04 | 79.16 | 91.84 | NGC: A Unified Framework for Learning with Open-World Noisy Data | - |