ReFACT: Updating Text-to-Image Models by Editing the Text Encoder

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Arad, Dana, Orgad, Hadas, Belinkov, Yonatan
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911868933636096
author Arad, Dana
Orgad, Hadas
Belinkov, Yonatan
author_facet Arad, Dana
Orgad, Hadas
Belinkov, Yonatan
contents Our world is marked by unprecedented technological, global, and socio-political transformations, posing a significant challenge to text-to-image generative models. These models encode factual associations within their parameters that can quickly become outdated, diminishing their utility for end-users. To that end, we introduce ReFACT, a novel approach for editing factual associations in text-to-image models without relaying on explicit input from end-users or costly re-training. ReFACT updates the weights of a specific layer in the text encoder, modifying only a tiny portion of the model's parameters and leaving the rest of the model unaffected. We empirically evaluate ReFACT on an existing benchmark, alongside a newly curated dataset. Compared to other methods, ReFACT achieves superior performance in both generalization to related concepts and preservation of unrelated concepts. Furthermore, ReFACT maintains image generation quality, making it a practical tool for updating and correcting factual information in text-to-image models.
format Preprint
id arxiv_https___arxiv_org_abs_2306_00738
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle ReFACT: Updating Text-to-Image Models by Editing the Text Encoder
Arad, Dana
Orgad, Hadas
Belinkov, Yonatan
Computation and Language
Computer Vision and Pattern Recognition
68T50
I.2.7
Our world is marked by unprecedented technological, global, and socio-political transformations, posing a significant challenge to text-to-image generative models. These models encode factual associations within their parameters that can quickly become outdated, diminishing their utility for end-users. To that end, we introduce ReFACT, a novel approach for editing factual associations in text-to-image models without relaying on explicit input from end-users or costly re-training. ReFACT updates the weights of a specific layer in the text encoder, modifying only a tiny portion of the model's parameters and leaving the rest of the model unaffected. We empirically evaluate ReFACT on an existing benchmark, alongside a newly curated dataset. Compared to other methods, ReFACT achieves superior performance in both generalization to related concepts and preservation of unrelated concepts. Furthermore, ReFACT maintains image generation quality, making it a practical tool for updating and correcting factual information in text-to-image models.
title ReFACT: Updating Text-to-Image Models by Editing the Text Encoder
topic Computation and Language
Computer Vision and Pattern Recognition
68T50
I.2.7
url https://arxiv.org/abs/2306.00738