Skip to content
The project page of paper: Universal Perturbation Attack Against Image Retrieval [ICCV 2019]
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

This is the project page of our paper:

Universal Perturbation Attack Against Image Retrieval,
Li, J., Ji, R., Liu, H., Hong, X., Gao, Y., & Tian, Q. ICCV 2019. [PDF]


Our codes are based on filipradenovic/cnnimageretrieval-pytorch (Commit c4fca89). Please refer to their repository for details.

The attack codes locate in cirtorch/examples/attack.

Prepare Features

  1. Follow the steps in filipradenovic/cnnimageretrieval-pytorch to download datasets and train the retrieval models.
  2. Refer to the function cluster() in cirtorch/examples/attack/myutil/ about extracting features and clustering.

Train Classifiers (Optional)

python -m PATH

Generate UAP

Refer to arguments in cirtorch/examples/attack/ for details.

Ranking Distillation

  1. Refer to cirtorch/examples/attack/ for extracting ranking list.
  2. Refer to cirtorch/examples/attack/ for distillation.

Typos in Paper

  1. Eq. 6 should be
  2. Eq. 7 should be

I'm sorry for typos in this paper. If you find more typos, please do not hesitate to point out in issues.


If our paper helps your research, please cite it in your publications:

author = {Li, Jie and Ji, Rongrong and Liu, Hong and Hong, Xiaopeng and Gao, Yue and Tian, Qi},
title = {Universal Perturbation Attack Against Image Retrieval},
booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
month = {October},
year = {2019}

Feel free to contact to the authors ( or create a new issue if you find any problems.

You can’t perform that action at this time.