Image Classification on CIFAR-10 (test) (IR-NR Robustness Error Rates)
4.9Top-1 Error Rate (IR-NR 1-0.0)lambda_A,W,S (AutoDO)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| lambda_A,W,S (AutoDO)Backbone=Wide-ResNet-28-10, Configuration=per-point augmentation, loss reweighting, and soft-labeling2021.03 | 4.9 | 11.8 | 5.8 | 13.1 | |
| FAABackbone=Wide-ResNet-28-102021.03 | 5 | 13.3 | 8.2 | 17.9 | |
| RAABackbone=Wide-ResNet-28-102021.03 | 5.2 | 13.4 | 7.8 | 17.4 | |
| lambda_A (AutoDO)Backbone=Wide-ResNet-28-10, Configuration=per-point augmentation hyperparameters2021.03 | 5.2 | 13 | 7.1 | 16.6 | |
| lambda_A,W (AutoDO)Backbone=Wide-ResNet-28-10, Configuration=per-point augmentation and loss reweighting2021.03 | 5.4 | 12.9 | 6.8 | 15.1 | |
| DADABackbone=Wide-ResNet-28-102021.03 | 5.5 | 14.2 | 10.6 | 20.2 | |
| lambda_ASHA (AutoDO)Backbone=Wide-ResNet-28-10, Configuration=shared augmentation hyperparameters2021.03 | 5.9 | 17.1 | 7.6 | 19.4 | |
| BaselineBackbone=Wide-ResNet-28-10, Augmentation=Standard (cropping, flipping, erasing)2021.03 | 6 | 16.3 | 12.8 | 22.8 |