A pytorch re-implementation for paper "Towards Deep Learning Models Resistant to Adversarial Attacks"
- pytorch>0.4
- torchvision
- tensorboardX
All the parameters are defined in config.py
exp_name
: experiment name, will be used for construct output directorysnap_dir
: root directory to save snapshots, it works withexp_name
to form a directory for a specific experiment
python train.py