k8xu/imageattribution

Detecting Image Attribution for Text-to-Image Diffusion Models in RGB and Beyond

Detecting Image Attribution for Text-to-Image Diffusion Models in RGB and Beyond

Paper | arXiv | Demo | Bibtex

Katherine Xu$^{1}$, Lingzhi Zhang$^{2}$, Jianbo Shi$^1$
$^1$ University of Pennsylvania, $^2$ Adobe Inc.

🚀 Updates

• 4/23/2024: Released a demo using EfficientFormer

Setup

• Clone this repo

git clone https://github.com/k8xu/ImageAttribution.git

• Install dependencies

conda create --name attribution python=3.10 -y
conda activate attribution
pip install opencv-python torch pillow

Demo

Download a torchscript checkpoint and place it under the specified folder.

Model Name	Torchscript	Folder	Test Accuracy
efficientformer	efficientformer_torchscript (118M)	./deployment/efficientformer	90.03%

We randomly sampled 10 test images per class, so you can quickly try our image attributor. Please check out ./images.

python demo.py --img {IMAGE PATH}

Citation

If you find our work useful, please cite our paper:

@article{xu2024detecting,
    title={Detecting Image Attribution for Text-to-Image Diffusion Models in RGB and Beyond},
    author={Xu, Katherine and Zhang, Lingzhi and Shi, Jianbo},
    journal={arXiv preprint arXiv:2403.19653},
    year={2024}
}

imageattribution's Issues

Demo model

Hi,
Thanks for sharing the model.
Is this model the image attributor or the model trained on the original RGB image?

Train/Test Dataset

Hi,
Thank you for this work. Will the train and test dataset comprising of synthetic images be made available to the public?