A PyTorch implementation of CVPR2020 paper Adversarial examples improve image recognition by Xie C, Tan M, Gong B, et al.
Thanks for guidance from Cihang Xie and Yingwei Li. The code is adapted from https://github.com/bearpaw/pytorch-classification.
This project is developed and tested in the following environments.
- Ubuntu 18.04
- CUDA 10.2
- GTX 1070 TI
- Python 3.8.1
python train.py
- train: training the model tune the parameters: iterations, epsilon,etc.
- net_rectified: ResNet architecture for advprop
- split_bn: split the normal BN and auxiliary BN to two separate Resnet34 model
- compile_to_deepinversion: compile Resnet from "net" to "resnet_cifar"