Implementation of the ICML 2024 paper titled "PID: Prompt-Independent Data Protection Against Latent Diffusion Models" by Ang Li*, Yichuan Mo, Mingjie Li, Yisen Wang. Our paper is available at http://arxiv.org/abs/2406.15305.
TODO:
- Initialize the repo. (2024/5/28)
- Training scripts & implementation of the PID.
- Implementation of the evaluation code.
- Visualizations & training data.
- [] Implementation of the baselines (FSGM, ASPL, AdvDM).
The minimum implementation of the PID is given in PID.py. Besides, we place clean images from the CelebA-HQ dataset in ./data/clean_images and the images already protected by PID in ./data/PID_images
sh PID.shFine-tuning can be started with a one-line command. Feel free to experiment with different training configurations.
sh train_dreambooth.sh # DreamBooth
sh train_drembooth_lora.sh # LoRAImplementation of the metrics used in the paper can be found in evaluate.py.
sh evaluate.shWant to have a discussion with the authors? Please open issues or send emails to charles_li@stu.pku.edu.cn
Please consider citing our work if you find it helpful!
This repo uses some of the code from the links below. We sincerely admire their great work!
- https://huggingface.co/docs/diffusers/en/training/dreambooth
- https://huggingface.co/docs/diffusers/en/training/lora
- https://github.com/VinAIResearch/Anti-DreamBooth/tree/main
- https://github.com/psyker-team/mist
- https://github.com/chaofengc/IQA-PyTorch
- https://github.com/timesler/facenet-pytorch
- https://github.com/krshrimali/No-Reference-Image-Quality-Assessment-using-BRISQUE-Model/tree/master/Python