Overview

This repository is a PyTorch implementation of the paper "GAMA: Generative Adversarial Multi-Object Scene Attacks" (NeurIPS'22).

Project page

Usage

1. Download the two folders from here and place them in classifer_models folders.
2. Install the packages listed in requirements.txt. Creating a conda environment is recommended.
3. To train a perturbation generator, run the following command:

python train.py --surr_model_type <surrogate model name> --data_name <voc/coco> --train_dir <path to dataset> --eps <l_infty noise strength> --batch_size 8 --epochs 20 --save_folder <path to trained models folder> --clip_backbone <clip model type> | tee <experiment name>.txt

4. To evaluate a trained perturbation generator, run the following command:

python eval.py --data_name <voc/coco> --gen_path <path to trained generator file (.pth)> 

Citing this work

If you find this work is useful in your research, please consider citing:

@inproceedings{
aich2022gama,
title={{GAMA}: Generative Adversarial Multi-Object Scene Attacks},
author={Abhishek Aich and Calvin-Khang Ta and Akash A Gupta and Chengyu Song and Srikanth Krishnamurthy and M. Salman Asif and Amit Roy-Chowdhury},
booktitle={Thirty-Sixth Conference on Neural Information Processing Systems},
year={2022},
url={https://openreview.net/forum?id=DRckHIGk8qw}
}

Contact

Please contact the first author of this paper - Abhishek Aich (aaich001@ucr.edu) for any further queries.

Acknowledgement

We thank the authors of the following repositories for making their code open-source.

1. https://github.com/megvii-research/ML-GCN
2. https://github.com/mingming97/multilabel-cam
3. https://github.com/Alibaba-AAIG/Beyond-ImageNet-Attack