Saved in:
Bibliographic Details
Main Authors: Li, Jiaqi, Wang, Guangming, Zheng, Shuntian, Ni, Minzhe, Lu, Xiaoman, Ye, Guanghui, Guan, Yu
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.21078
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908953347096576
author Li, Jiaqi
Wang, Guangming
Zheng, Shuntian
Ni, Minzhe
Lu, Xiaoman
Ye, Guanghui
Guan, Yu
author_facet Li, Jiaqi
Wang, Guangming
Zheng, Shuntian
Ni, Minzhe
Lu, Xiaoman
Ye, Guanghui
Guan, Yu
contents Temporal Action Localization (TAL) requires identifying both the boundaries and categories of actions in untrimmed videos. While vision-language models (VLMs) offer rich semantics to complement visual evidence, existing approaches tend to overemphasize linguistic priors at the expense of visual performance, leading to a pronounced modality bias. We propose ActionVLM, a vision-language aggregation framework that systematically mitigates modality bias in TAL. Our key insight is to preserve vision as the dominant signal while adaptively exploiting language only when beneficial. To this end, we introduce (i) a debiasing reweighting module that estimates the language advantage-the incremental benefit of language over vision-only predictions-and dynamically reweights language modality accordingly, and (ii) a residual aggregation strategy that treats language as a complementary refinement rather than the primary driver. This combination alleviates modality bias, reduces overconfidence from linguistic priors, and strengthens temporal reasoning. Experiments on THUMOS14 show that our model outperforms state-of-the-art by up to 3.2% mAP. Our code is available at https://github.com/JiaqiLi404/ActionVLM
format Preprint
id arxiv_https___arxiv_org_abs_2601_21078
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Towards Mitigating Modality Bias in Vision-Language Models for Temporal Action Localization
Li, Jiaqi
Wang, Guangming
Zheng, Shuntian
Ni, Minzhe
Lu, Xiaoman
Ye, Guanghui
Guan, Yu
Computer Vision and Pattern Recognition
Temporal Action Localization (TAL) requires identifying both the boundaries and categories of actions in untrimmed videos. While vision-language models (VLMs) offer rich semantics to complement visual evidence, existing approaches tend to overemphasize linguistic priors at the expense of visual performance, leading to a pronounced modality bias. We propose ActionVLM, a vision-language aggregation framework that systematically mitigates modality bias in TAL. Our key insight is to preserve vision as the dominant signal while adaptively exploiting language only when beneficial. To this end, we introduce (i) a debiasing reweighting module that estimates the language advantage-the incremental benefit of language over vision-only predictions-and dynamically reweights language modality accordingly, and (ii) a residual aggregation strategy that treats language as a complementary refinement rather than the primary driver. This combination alleviates modality bias, reduces overconfidence from linguistic priors, and strengthens temporal reasoning. Experiments on THUMOS14 show that our model outperforms state-of-the-art by up to 3.2% mAP. Our code is available at https://github.com/JiaqiLi404/ActionVLM
title Towards Mitigating Modality Bias in Vision-Language Models for Temporal Action Localization
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2601.21078