Noisy-label detection on MNIST
99.83Accuracy2D geometric metric
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| 2D geometric metricBackbone=Resnet50, Label error (%)=10%2026.06 | 99.83 | 98.89 | 99.45 | |
| 3D multi-metricBackbone=Resnet50, Label error (%)=10%2026.06 | 99.82 | 98.63 | 99.57 | |
| 2D geometric metricBackbone=Resnet50, Label error (%)=20%2026.06 | 99.77 | 99.44 | 99.4 | |
| 3D multi-metricBackbone=Resnet50, Label error (%)=20%2026.06 | 99.77 | 99.42 | 99.42 | |
| 2D geometric metricBackbone=Resnet50, Label error (%)=40%2026.06 | 99.69 | 99.58 | 99.64 | |
| 3D multi-metricBackbone=Resnet50, Label error (%)=40%2026.06 | 99.69 | 99.58 | 99.65 | |
| CleanlabLabel error (%)=5%2026.06 | 99.62 | 95.52 | 96.86 | |
| CleanlabLabel error (%)=10%2026.06 | 99.57 | 97.66 | 98.07 | |
| 2D geometric metricBackbone=Vit, Label error (%)=20%2026.06 | 99.49 | 97.77 | 99.72 | |
| CleanlabLabel error (%)=20%2026.06 | 99.35 | 98.57 | 98.18 | |
| 2D geometric metricBackbone=Vit, Label error (%)=40%2026.06 | 99.28 | 98.52 | 99.7 | |
| CleanlabLabel error (%)=40%2026.06 | 99.1 | 99.13 | 98.62 | |
| 2D geometric metricBackbone=Resnet50, Label error (%)=5%2026.06 | 96.55 | 59.16 | 99.83 | |
| 3D multi-metricBackbone=Vit, Label error (%)=40%2026.06 | 96.55 | 92.07 | 99.97 | |
| 3D multi-metricBackbone=Resnet50, Label error (%)=5%2026.06 | 93.2 | 42.34 | 99.97 | |
| 2D geometric metricBackbone=Vit, Label error (%)=5%2026.06 | 92.7 | 40.61 | 99.97 | |
| 2D geometric metricBackbone=Vit, Label error (%)=10%2026.06 | 92.42 | 56.86 | 99.98 | |
| 3D multi-metricBackbone=Vit, Label error (%)=20%2026.06 | 90.2 | 67.12 | 100 | |
| 3D multi-metricBackbone=Vit, Label error (%)=5%2026.06 | 89.32 | 31.85 | 99.97 | |
| 3D multi-metricBackbone=Vit, Label error (%)=10%2026.06 | 88.91 | 47.39 | 100 |