Saved in:
Bibliographic Details
Main Authors: Wu, Xiangyang, Liu, Liu, Yu, Baosheng, Qiu, Jiayan, Shi, Zhenwei
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2511.08238
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911263009800192
author Wu, Xiangyang
Liu, Liu
Yu, Baosheng
Qiu, Jiayan
Shi, Zhenwei
author_facet Wu, Xiangyang
Liu, Liu
Yu, Baosheng
Qiu, Jiayan
Shi, Zhenwei
contents Vision-language fine-tuning has emerged as an efficient paradigm for constructing multimodal foundation models. While textual context often highlights semantic relationships within an image, existing fine-tuning methods typically overlook this information when aligning vision and language, thus leading to suboptimal performance. Toward solving this problem, we propose a method that can improve multimodal alignment and fusion based on both semantics and relationships.Specifically, we first extract multilevel semantic features from different vision encoder to capture more visual cues of the relationships. Then, we learn to project the vision features to group related semantics, among which are more likely to have relationships. Finally, we fuse the visual features with the textual by using inheritable cross-attention, where we globally remove the redundant visual relationships by discarding visual-language feature pairs with low correlation. We evaluate our proposed method on eight foundation models and two downstream tasks, visual question answering and image captioning, and show that it outperforms all existing methods.
format Preprint
id arxiv_https___arxiv_org_abs_2511_08238
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Remodeling Semantic Relationships in Vision-Language Fine-Tuning
Wu, Xiangyang
Liu, Liu
Yu, Baosheng
Qiu, Jiayan
Shi, Zhenwei
Computer Vision and Pattern Recognition
Artificial Intelligence
Vision-language fine-tuning has emerged as an efficient paradigm for constructing multimodal foundation models. While textual context often highlights semantic relationships within an image, existing fine-tuning methods typically overlook this information when aligning vision and language, thus leading to suboptimal performance. Toward solving this problem, we propose a method that can improve multimodal alignment and fusion based on both semantics and relationships.Specifically, we first extract multilevel semantic features from different vision encoder to capture more visual cues of the relationships. Then, we learn to project the vision features to group related semantics, among which are more likely to have relationships. Finally, we fuse the visual features with the textual by using inheritable cross-attention, where we globally remove the redundant visual relationships by discarding visual-language feature pairs with low correlation. We evaluate our proposed method on eight foundation models and two downstream tasks, visual question answering and image captioning, and show that it outperforms all existing methods.
title Remodeling Semantic Relationships in Vision-Language Fine-Tuning
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2511.08238