README.md

Attentive Eraser: Unleashing Diffusion Model’s Object Removal Potential via Self-Attention Redirection Guidance (AAAI 2025)

Introduction

Attentive Eraser is a novel tuning-free method that enhances object removal capabilities in pre-trained diffusion models. This official implementation demonstrates the method's efficacy, leveraging altered self-attention mechanisms to prioritize background over foreground in the image generation process.

Downloading Pretrained Diffusion Models

The pretrained diffusion models can be downloaded from the link below for offline loading.
SDXL: https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0
SD2.1: https://huggingface.co/stabilityai/stable-diffusion-2-1-base

Getting Started

git clone https://github.com/Anonym0u3/AttentiveEraser.git
cd AttentiveEraser
conda create -n AE python=3.9
conda activate AE
pip install -r requirements.txt
# run SDXL+SIP
python main.py

More experimental versions can be found in the notebook folder.

Citation

If you find this project useful in your research, please consider citing it:

@inproceedings{sun2025attentive,
  title={Attentive Eraser: Unleashing Diffusion Model’s Object Removal Potential via Self-Attention Redirection Guidance},
  author={Sun, Wenhao and Cui, Benlei and Dong, Xue-Mei and Tang, Jingqun},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2025}
}

Acknowledgments

This repository is built upon and utilizes the following repositories:

• MasaCtrl
• diffusers

We would like to express our sincere thanks to the authors and contributors of these repositories for their incredible work, which greatly enhanced the development of this repository.