Unified Concept Editing in Diffusion Models

By Rohit Gandikota, Hadas Orgad, Yonatan Belinkov, Joanna Materzy'nska, David Bau. IEEE Workshop/Winter Conference on Applications of Computer Vision, 2023

425 citations recorded by the source corpus.

Text-to-image models suffer from various safety issues that may limit their suitability for deployment. Previous methods have separately addressed individual issues of bias, copyright, and offensive content in text-to-image models. However, in the real world, all of these issues appear simultaneously in the same model. We present a method that tackles all issues with a single approach. Our method, Unified Concept Editing (UCE), edits the model without training using a closed-form solution, and scales seamlessly to concurrent edits on text-conditional diffusion models.We present scalable simultaneous debiasing, style erasure, and content moderation by editing text-to-image projections, and perform extensive experiments demonstrating improved efficacy and scalability over prior work. Our code is available at unified.baulab.info.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/e367992ca8deed39d18717a3ef08aa05a94128c0

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Yanzhu Liu
2. Adams Wai-Kin Kong
3. Leyang Li
4. Zilan Wang
5. Shilin Lu

Citations in the corpus

• MACE: Mass Concept Erasure in Diffusion Models (2024, methodology): “UCE [17] proves more effective, but its specificity decreases rapidly when more than 10 concepts are erased.”