Code for the paper Generalizable Data-free Objective for Crafting Universal Adversarial Perturbations
Accepted in IEEE Transactions on Pattern Analysis and Machine Intelligence!
Mopuri Konda Reddy,Aditya Ganeshan, R. Venkatesh Babu
This repository contains code to craft and evaluate GD-UAP on the following task:
1) Classification.
2) Segmentation.
3) Depth Estimation.
Please look into each of the folders for usage instructions.
Perturbations crafted using the proposed algorithm are provided in this link. After extracting them, and placing them in the respective folders (In each task), you can use the evaluation code provided in each task for evaluation.
@article{gduap-mopuri-2018,
title={Generalizable Data-free Objective for Crafting Universal Adversarial
Perturbations},
author={Mopuri, Konda Reddy and Ganeshan, Aditya and Babu, R Venkatesh},
booktitle = {arXiv preprint arXiv: 1801.08092 },
year = {2018}
}