Receler: Reliable Concept Erasing of Text-to-Image Diffusion Models via Lightweight Erasers

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Huang, Chi-Pin, Chang, Kai-Po, Tsai, Chung-Ting, Lai, Yung-Hsuan, Yang, Fu-En, Wang, Yu-Chiang Frank
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911959932207104
author Huang, Chi-Pin
Chang, Kai-Po
Tsai, Chung-Ting
Lai, Yung-Hsuan
Yang, Fu-En
Wang, Yu-Chiang Frank
author_facet Huang, Chi-Pin
Chang, Kai-Po
Tsai, Chung-Ting
Lai, Yung-Hsuan
Yang, Fu-En
Wang, Yu-Chiang Frank
contents Concept erasure in text-to-image diffusion models aims to disable pre-trained diffusion models from generating images related to a target concept. To perform reliable concept erasure, the properties of robustness and locality are desirable. The former refrains the model from producing images associated with the target concept for any paraphrased or learned prompts, while the latter preserves its ability in generating images with non-target concepts. In this paper, we propose Reliable Concept Erasing via Lightweight Erasers (Receler). It learns a lightweight Eraser to perform concept erasing while satisfying the above desirable properties through the proposed concept-localized regularization and adversarial prompt learning scheme. Experiments with various concepts verify the superiority of Receler over previous methods.
format Preprint
id arxiv_https___arxiv_org_abs_2311_17717
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Receler: Reliable Concept Erasing of Text-to-Image Diffusion Models via Lightweight Erasers
Huang, Chi-Pin
Chang, Kai-Po
Tsai, Chung-Ting
Lai, Yung-Hsuan
Yang, Fu-En
Wang, Yu-Chiang Frank
Computer Vision and Pattern Recognition
Machine Learning
Concept erasure in text-to-image diffusion models aims to disable pre-trained diffusion models from generating images related to a target concept. To perform reliable concept erasure, the properties of robustness and locality are desirable. The former refrains the model from producing images associated with the target concept for any paraphrased or learned prompts, while the latter preserves its ability in generating images with non-target concepts. In this paper, we propose Reliable Concept Erasing via Lightweight Erasers (Receler). It learns a lightweight Eraser to perform concept erasing while satisfying the above desirable properties through the proposed concept-localized regularization and adversarial prompt learning scheme. Experiments with various concepts verify the superiority of Receler over previous methods.
title Receler: Reliable Concept Erasing of Text-to-Image Diffusion Models via Lightweight Erasers
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2311.17717