| SOMNet (ResNet-50) | 65.0% | 85.1% | 72.2% | 89.5% | Learning from Web Data with Self-Organizing Memory Module | - |
| HAR (InceptionResNet-v2) | 67.1% | 86.7% | 75.0% | 90.6% | Heteroskedastic and Imbalanced Deep Learning with Adaptive Regularization | |
| MoPro (ResNet-50) | 67.8% | 87.0% | 73.9% | 90.0% | MoPro: Webly Supervised Learning with Momentum Prototypes | |
| CurriculumNet (Inception-v2） | 64.8% | 84.9% | 72.1% | 89.2% | CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images | |
| MentorNet (InceptionResNet-V2) | 62.5% | 83.0% | 70.8% | 88.0% | MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels | |
| MentorMix (InceptionResNet-V2) | 67.5% | 87.2% | 74.3% | 90.5% | Beyond Synthetic Noise: Deep Learning on Controlled Noisy Labels | - |
| CurriculumNet (InceptionResNet-v2) | - | - | 79.3% | 93.6% | CurriculumNet: Weakly Supervised Learning from Large-Scale Web Images | |
| NCR+Mixup+DA (ResNet-50) | - | - | 76.8 | - | Learning with Neighbor Consistency for Noisy Labels | |
| Heteroscedastic (InceptionResNet-v2) | - | - | 76.6% | 92.1% | Correlated Input-Dependent Label Noise in Large-Scale Image Classification | - |
| ERM + CMW-Net-SL (InceptionResNet-V2) | - | - | 77.9% | 92.6% | CMW-Net: Learning a Class-Aware Sample Weighting Mapping for Robust Deep Learning | |
| SCC (ResNet50-D) | 70.66% | 88.46% | 75.78% | 91.07% | Webly Supervised Image Classification with Self-Contained Confidence | |