Anisotropic Modality Align

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yu, Xiaomin, Li, Yijiang, Zhang, Yuhui, Zhao, Hanzhen, Yang, Yue, Tang, Hao, Song, Yue, Hu, Xiaobin, Qin, Chengwei, Yan, Shuicheng, Xiong, Hui
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911662833926144
author Yu, Xiaomin
Li, Yijiang
Zhang, Yuhui
Zhao, Hanzhen
Yang, Yue
Tang, Hao
Song, Yue
Hu, Xiaobin
Qin, Chengwei
Yan, Shuicheng
Xiong, Hui
author_facet Yu, Xiaomin
Li, Yijiang
Zhang, Yuhui
Zhao, Hanzhen
Yang, Yue
Tang, Hao
Song, Yue
Hu, Xiaobin
Qin, Chengwei
Yan, Shuicheng
Xiong, Hui
contents Training multimodal large language models has long been limited by the scarcity of high-quality paired multimodal data. Recent studies show that the shared representation space of pretrained multimodal contrastive models can serve as a bridge, enabling models to perform multimodal training with unimodal data. However, the key premise of this paradigm remains insufficiently understood: can representations from different modalities be reliably interchanged? The core obstacle lies in the persistent Modality Gap in the shared space. In this work, we revisit the geometric nature of the modality gap. We find that modality representations already share compatible dominant semantic geometry. What truly hinders modality interchangeability is not a simple global shift, but an anisotropic residual structure concentrated along a small number of dominant directions. Based on this finding, we further propose the principle of anisotropic modality gap alignment: effective modality alignment should align with the target-modality distribution while preserving the semantic structure of the source modality. Guided by this principle, we propose an anisotropic geometric correction framework, AnisoAlign, for unpaired modality alignment. This framework leverages the internal geometric prior of the target modality and performs bounded correction on source-modality representations, thereby constructing substitute representations in the target modality. Experiments confirm its benefits in both geometric diagnostics and text-only MLLM training. Overall, this work recasts the modality gap from an empirical observation into a correctable, structured geometric phenomenon and provides a new representation alignment perspective for training multimodal models with unimodal data.
format Preprint
id arxiv_https___arxiv_org_abs_2605_07825
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Anisotropic Modality Align
Yu, Xiaomin
Li, Yijiang
Zhang, Yuhui
Zhao, Hanzhen
Yang, Yue
Tang, Hao
Song, Yue
Hu, Xiaobin
Qin, Chengwei
Yan, Shuicheng
Xiong, Hui
Multimedia
Computer Vision and Pattern Recognition
Training multimodal large language models has long been limited by the scarcity of high-quality paired multimodal data. Recent studies show that the shared representation space of pretrained multimodal contrastive models can serve as a bridge, enabling models to perform multimodal training with unimodal data. However, the key premise of this paradigm remains insufficiently understood: can representations from different modalities be reliably interchanged? The core obstacle lies in the persistent Modality Gap in the shared space. In this work, we revisit the geometric nature of the modality gap. We find that modality representations already share compatible dominant semantic geometry. What truly hinders modality interchangeability is not a simple global shift, but an anisotropic residual structure concentrated along a small number of dominant directions. Based on this finding, we further propose the principle of anisotropic modality gap alignment: effective modality alignment should align with the target-modality distribution while preserving the semantic structure of the source modality. Guided by this principle, we propose an anisotropic geometric correction framework, AnisoAlign, for unpaired modality alignment. This framework leverages the internal geometric prior of the target modality and performs bounded correction on source-modality representations, thereby constructing substitute representations in the target modality. Experiments confirm its benefits in both geometric diagnostics and text-only MLLM training. Overall, this work recasts the modality gap from an empirical observation into a correctable, structured geometric phenomenon and provides a new representation alignment perspective for training multimodal models with unimodal data.
title Anisotropic Modality Align
topic Multimedia
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2605.07825