AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

LightAVSeg: Lightweight Audio-Visual Segmentation

Qing Zhong, Guodong Ding, Lingqiao Liu, Zaiwen Feng, Lin Yuanbo Wu, Angela Yao

ICML 2026 · Vision-language and multimodal › Cross-modal learning · approach: Audio-visual source localization · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Audio-Guided Visual Enhancement (AGVE) (ablation)static-concatenation baseline · AVSBench MS3M_J: 44.4 -> 46.8Improved
add Cross-Modal Fusion Decoder (CMFD) to AGVE (ablation)AGVE · AVSBench MS3M_J: 46.8 -> 48.3Improved
add Hierarchical Audio Refinement (HAR) to AGVE (ablation)AGVE · AVSBench MS3M_J: 46.8 -> 49.3Improved
add CMFD and L_msa to full encoder (ablation)AGVE+HAR · AVSBench MS3M_J: 49.3 -> 50.4Improved
add L_msa auxiliary loss (ablation)L_seg · AVSBench MS3M_J: 49.3 -> 50.4 (+1.1 mIoU)Improved
add L_msa auxiliary loss (ablation)L_seg · AVSBench MS3M_F: 60.8 -> 62.6 (+1.8)Improved
add L_AVM auxiliary loss (ablation)L_seg · AVSBench MS3M_J: 49.3 -> 49.2No clear effect
add L_mix auxiliary loss (ablation)L_seg · AVSBench MS3M_J: 49.3 -> 48.8Hurt
use SeaFormer-Tiny instead of SeaFormer-Large visual backboneLightAVSeg · AVSBench MS3M_J: 50.4 -> 30.6Hurt
use SeaFormer-Base instead of SeaFormer-Large visual backboneLightAVSeg · AVSBench MS3M_J: 50.4 -> 44.2Hurt
use ResNet-50 instead of SeaFormer-Large visual backboneLightAVSeg · AVSBench MS3M_J: 50.4 -> 52.9; latency 163.4 -> 675.0 msMixed
use YamNet instead of MobileNetV2 audio backboneLightAVSeg · AVSBench MS3M_J: 50.4 -> 47.8Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy