SemanticAdv (ECCV 2020)

This is official PyTorch implementation of ECCV 2020 paper SemanticAdv: Generating Adversarial Examplesvia Attribute-conditioned Image Editing by Haonan Qiu, Chaowei Xiao, Lei Yang, Xinchen Yan, Honglak Lee, Bo Li.

Please follow the instructions to run the code.

Overview

--attacks    #core function for SemanticAdv
--Face       #demo on face (CelebA) verification task
--Street     #demo on street (Cityscapes) segmentation task

Core Function (embed it in your own model)

from attacks import semantic_attack

adversary = semantic_attack.FP_CW_TV(learning_rate, 
                                     maximal_iteration,
                                     lambda_for_tv_loss,
                                     threshold) 

edit_final, adv_loss, tv_loss = adversary(G_dec=decoder,
                                          emb1=feature_map1,
                                          emb2=feature_map2,
                                          model=target_model,
                                          loss_func=loss_function,
                                          target_label=target_label,
                                          targeted=True)

test environment

pytorch==1.1.0
torchvision==0.3.0
scipy==1.2.1
pillow=6.1.0
dominate=2.4.0
dill=0.3.1.1
scikit-image=0.16.2

Run Face Demo

cd Face
bash scripts/download_pretrained_face_model.sh
python verification_attack.py
python verification_attack_untarget.py

Run the following lines to replicate results in the ECCV paper.

bash scripts/download_all_aligned_images.sh
python attack_paper_reproduction.py

reference repo

generative model (StarGAN)
https://github.com/yunjey/stargan

verification model
https://github.com/yl-1993/hfsoftmax

Run Street Demo

cd Street
bash scripts/download_pretrained_mask2image_city.sh
bash scripts/download_pretrained_seg_model.sh
bash scripts/attack_seg_p11.sh

reference repo

generative model
https://github.com/xcyan/neurips18_hierchical_image_manipulation

segmentation model (DRN)
https://github.com/fyu/drn

Citation

If you find this useful, please cite our work as follows:

@inproceedings{qiu2019semanticadv,
  title={Semanticadv: Generating adversarial examples via attribute-conditioned image editing},
  author={Qiu, Haonan and Xiao, Chaowei and Yang, Lei and Yan, Xinchen and Lee, Honglak and Li, Bo},
  booktitle={ECCV},
  year={2020}
}

Acknowledgements

We would like to thank the amazing developers and the open-sourcing community. Our implementation has especially been benefited from the following excellent repositories:

• StarGAN Model: https://github.com/yunjey/stargan
• Face Verification Model: https://github.com/yl-1993/hfsoftmax
• Hierachical Image Editing: https://github.com/xcyan/neurips18_hierchical_image_manipulation
• DRN Segmentation Model: https://github.com/fyu/drn