Hiding Visual Information via Obfuscating Adversarial Perturbations

This repository is an implementation of the ICCV 2023 paper "Hiding Visual Information via Obfuscating Adversarial Perturbations".

Environment settings and libraries we used in our experiments

This project is tested under the following environment settings:

• OS: Ubuntu 18.04.6
• Cuda: 11.1, Cudnn: v8.05
• Python: 3.7.12
• TensorFlow: 1.9.0
• PyTorch: >= 1.10.2
• Torchvision: >= 0.11.3

Preparation

1. Align and crop the face image to $112 \times 112$ and put it under data/align.
2. Put the pre-trained generative model (key model) under the checkpoints. We provide a pre-trained model for testing in face recognition tasks.

Image Encryption for Face Recognition

Running command to encrypt the image:

python main.py --model pix2pix --task face_recognition

Image Encryption for Classification

Running command to encrypt the image:

python main.py --model pix2pix --task classification --src_model Resnet18 --batch_size 10 --root ./data/cif

Acknowledgements

This code is built on CycleGAN, TIP-IM and MS-SSIM. We thank the authors for sharing the codes.