This is the code for the paper: Combining Adversaries with Anti-adversaries in Training
The requiring environment is as bellow:
- Linux
- python 3.8
- pytorch 1.9.0
- torchvision 0.10.0
Here are two examples for training imbalanced and noisy data:
ResNet32 on CIFAR10-LT with imbalanced factor of 10:
python CAAT.py --dataset cifar10 --imbalanced_factor 10
ResNet32 on noisy CIFAR10 with 20% pair-flip noise:
python CAAT.py --dataset cifar10 --corruption_type flip2 --corruption_ratio 0.2