Image Classification on CIFAR-100 40% asymmetric label noise
77.78Final Classification AccuracyC2D (DivideMix with SimCLR)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| C2D (DivideMix with SimCLR)Architecture=ResNet-502021.03 | 77.78 | 77.92 | |
| C2D (ELR+ with SimCLR)Architecture=ResNet-342021.03 | 77.63 | 77.87 | |
| ELR+Architecture=ResNet-342021.03 | 77.5 | — | |
| C2D (DivideMix with SimCLR)Architecture=PreAct ResNet-182021.03 | 75.06 | 75.48 | |
| DivideMixArchitecture=PreAct ResNet-182021.03 | 72.4 | 72.2 | |
| CE+mixup with SimCLRArchitecture=ResNet-342021.03 | 54.75 | 65.19 |