This repository is the source code for How to Trace Latent Generative Model Generated Images without Artificial Watermark? (ICML 2024).
See requirements.txt
The version of Python is 3.10
For example, generating images using Stable Diffusion v1-5:
python generate_samples_all.py --arch sd --save_dir ./sdv15_generated_imgs/Generating images using Stable Diffusion v2-base:
python generate_samples_all.py --arch sdv2base --save_dir ./sdv2base_generated_imgs/For example, to conduct reverse-engineering on the belonging images of Stable Diffusion Stable Diffusion v2-base:
python run_write_re_loss_to_txt.py --gpu 0 --filePath ./sdv2base_generated_imgs/ --model_type sdv2base --lr 0.05 \
--num_iter 100 --write_txt_path ./ReconstructionLosses_Model_sdv2base_Images_sdv2baseGenerated.txtTo conduct reverse-engineering on the non-belonging images (images generated by Stable Diffusion v1-5 here):
python run_write_re_loss_to_txt.py --gpu 0 --filePath ./sdv15_generated_imgs/ --model_type sdv2base --lr 0.05 \
--num_iter 100 --write_txt_path ./ReconstructionLosses_Model_sdv2base_Images_sdv15Generated.txtpython detection.py \
--txt_1 ./ReconstructionLosses_Model_sdv2base_Images_sdv2baseGenerated.txt \
--txt_2 ./ReconstructionLosses_Model_sdv2base_Images_sdv15Generated.txt \
--label_1 'Belongings' \
--label_2 'Non-belongings' \
--threshold 1e-4 \
--save_name ./distribution.pdfNote that the threshold needs to be adjusted for different inspected model based on the distribution of reconstruction losses on its belonging images.
You are encouraged to cite the following papers if you use the repo for academic research.
@inproceedings{wang2024trace,
title={How to Trace Latent Generative Model Generated Images without Artificial Watermark?},
author={Wang, Zhenting and Sehwag Vikash, Chen, Chen and Lyu, Lingjuan and Metaxas, Dimitris N and Ma, Shiqing},
booktitle={International Conference on Machine Learning},
year={2024}
}
@inproceedings{wang2023did,
title={Where Did I Come From? Origin Attribution of AI-Generated Images},
author={Wang, Zhenting and Chen, Chen and Zeng, Yi and Lyu, Lingjuan and Ma, Shiqing},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023}
}