Saved in:
Bibliographic Details
Main Authors: Zhong, Qing, Ding, Guodong, Liu, Lingqiao, Feng, Zaiwen, Wu, Lin Yuanbo, Yao, Angela
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.08805
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909029704400896
author Zhong, Qing
Ding, Guodong
Liu, Lingqiao
Feng, Zaiwen
Wu, Lin Yuanbo
Yao, Angela
author_facet Zhong, Qing
Ding, Guodong
Liu, Lingqiao
Feng, Zaiwen
Wu, Lin Yuanbo
Yao, Angela
contents Audio-Visual Segmentation (AVS) targets pixel level localization of sounding emitting objects in videos. However, existing models rely on dense cross-modal attention with quadratic computational cost, limiting their suitability for resource efficient deployment. Most efficiency oriented methods focus on backbone reduction and overlook the interaction module as the primary bottleneck. This paper proposes LightAVSeg, a lightweight framework that replaces heavy attention with a decoupled design for semantic filtering and spatial grounding, resulting in interaction costs that scale linearly with spatial resolution. Furthermore, we introduce an auxiliary alignment loss to enforce semantic consistency during training with zero inference overhead. Extensive experiments demonstrate that LightAVSeg achieves a new state-of-the-art among lightweight methods: with 20.5M parameters ~1/7 of AVSegFormer), it reaches 50.4 mIoU on the MS3 benchmark and enables efficient inference on a mobile processor.
format Preprint
id arxiv_https___arxiv_org_abs_2605_08805
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle LightAVSeg: Lightweight Audio-Visual Segmentation
Zhong, Qing
Ding, Guodong
Liu, Lingqiao
Feng, Zaiwen
Wu, Lin Yuanbo
Yao, Angela
Computer Vision and Pattern Recognition
Audio-Visual Segmentation (AVS) targets pixel level localization of sounding emitting objects in videos. However, existing models rely on dense cross-modal attention with quadratic computational cost, limiting their suitability for resource efficient deployment. Most efficiency oriented methods focus on backbone reduction and overlook the interaction module as the primary bottleneck. This paper proposes LightAVSeg, a lightweight framework that replaces heavy attention with a decoupled design for semantic filtering and spatial grounding, resulting in interaction costs that scale linearly with spatial resolution. Furthermore, we introduce an auxiliary alignment loss to enforce semantic consistency during training with zero inference overhead. Extensive experiments demonstrate that LightAVSeg achieves a new state-of-the-art among lightweight methods: with 20.5M parameters ~1/7 of AVSegFormer), it reaches 50.4 mIoU on the MS3 benchmark and enables efficient inference on a mobile processor.
title LightAVSeg: Lightweight Audio-Visual Segmentation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2605.08805