Multimodal Model Editing (MMED) aims to correct erroneous knowledge in multimodal models. Existing evaluation methods, adapted from textual model editing, overstate success by relying on low-similarity or random inputs, obscure overfitting. We propose a comprehensive locality evaluation framework, covering three key dimensions: random-image locality, no-image locality, and consistent-image locality, operationalized through seven distinct data types, enabling a detailed and structured analysis of multimodal edits. We introduce De-VQA, a dynamic evaluation for visual question answering, uncovering a phenomenon we term transient blindness, overfitting to edit-similar text while ignoring visuals. Token analysis shows edits disproportionately affect textual tokens. We propose locality-aware adversarial losses to balance cross-modal representations. Empirical results demonstrate that our approach consistently outperforms existing baselines, reducing transient blindness and improving locality by 17% on average.
@article{arxiv.2511.13243,
title = {Uncovering and Mitigating Transient Blindness in Multimodal Model Editing},
author = {Xiaoqi Han and Ru Li and Ran Yi and Hongye Tan and Zhuomin Liang and Víctor Gutiérrez-Basulto and Jeff Z. Pan},
journal= {arXiv preprint arXiv:2511.13243},
year = {2025}
}