python gdpa.py --dataset [imagenet|vggface] --data_path [FOLDER_NAME] 

If on VGGFace, please add --vgg_model_path [MODEL_PATH]

optional arguments:
  --patch_size            size of adversarial patch
  --alpha                 $\alpha$ in paper
  --beta                  $\beta$ in paper
  --exp                   exp name in logging
  --epochs                epochs for training
  --lr_gen                learning rate
  --batch_size            batch size
  --device                cuda or cpu

python gdpa_at.py --data_path [FOLDER_NAME] --vgg_model_path [MODEL_PATH] 

optional arguments:
  --patch_size            size of adversarial patch
  --beta                  $\beta$ in paper
  --lr_gen                learning rate for generator
  --lr_clf                learning rate for classifier
  --save_freq             frequency of saving the model
  --epochs                epochs for training
  --batch_size            batch size
  --device                cuda or cpu
  --enable_testing        testing during training

tensorboard --logdir logs/exp/gdpa/

tensorboard --logdir logs/exp/gdpa_at/

@article{xiang2021gdpa,
    title={Generative Dynamic Patch Attack},
    author={Xiang Li and Shihao Ji},
    journal={British Machine Vision Conference (BMVC)},
    year={2021}
}