This repository contains code for the paper:
- Mimic and Fool: A Task Agnostic Adversarial Attack. Akshay Chaturvedi and Utpal Garain. IEEE TNNLS, 2020. [pdf]
- Keras v2.2.4 with Tensorflow (v 1.12.0) backend
- numpy, opencv-python
The finetuned feature extractor weights can be downloaded from:
Show Attend and Tell: https://drive.google.com/open?id=1QmIze742K4aWTOuzhRF4nbiw5lHgN8TY
Extract the two files in their respective folders.
Under any setting, run prepare_target.py followed by main.py. The adversarial files will be saved in adv_imgs/ folder.