DEFENDING AGAINST PHYSICALLY REALIZABLE ATTACKS ON IMAGE CLASSIFICATION

This repository is reproduce code about the article DEFENDING AGAINST PHYSICAL REALIZABLE ATTACKS ON IMAGE CLASSIFICATION.

Original Paper: Wu, T., Tong, L., & Vorobeychik, Y. (2019). Defending Against Physically Realizable Attacks on Image Classification. arXiv preprint arXiv:1909.09552, Available: https://arxiv.org/pdf/1909.09552.pdf

Original Code: https://github.com/tongwu2020/phattacks

The code for sign can found in './Sign'. 
ROA.py is used to generate the attack images. 
train.py is used to train original model and DOA Model. 
The code of ROA with two algorithms can be found in model.py

The code for VGGface can found in './VGGface'. 

Dataset:

• VGGface and LISA can be found: https://github.com/tongwu2020/phattacks/releases/tag/Data%26Model

Pretrained_model:

• Weight of LISA DOA Model can be found: './Sign/model_DOA.pkl'
• Weight of LISA Original Model can be found: './Sign/model.pkl'
• Weight of VGG Model can be found: https://drive.google.com/file/d/1VXNEeivr_U9o2pf-YmWz4wgy2KoZvYlK/view?usp=sharing

Authors

• CHi Xu cx1g19@soton.ac.uk
• Yin Bao yb1a19@soton.ac.uk
• Zhaori Guo zg2n19@soton.ac.uk