AI Paper Atlas

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

Delayed Bidirectional Alignment via Disentangled Audio Semantics for Audio-Visual Segmentation

Jingqi Tian, Yiheng Du, Haoji Zhang, Yuji Wang, Isaac Ning Lee, Xulong Bai, Tianrui Zhu, Jingxuan Niu, Yansong Tang

ECCV 2026 · Vision-language and multimodal › Cross-modal learning · approach: Audio-visual source localization · 4 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 Query Module (prototype-bank queries) (ablation)baseline · AVS-MS3J&F: 69.71 -> 71.89Improved
add Audio Query Module (ablation)baseline · AVSSJ&F: 48.63 -> 49.80Improved
add Contrastive Optimization Module (ablation)baseline+AQM · AVS-MS3J&F: 71.89 -> 74.07Improved
add Contrastive Optimization Module (ablation)baseline+AQM · AVSSJ&F: 49.80 -> 51.70Improved
add delayed bidirectional alignment module (AVAM) (ablation)baseline · AVS-MS3J&F: 69.71 -> 73.16Improved
add AQM and COM on top of AVAM (ablation)AVAM · AVS-MS3J&F: 73.16 -> 76.01Improved
add AQM and COM on top of AVAM (ablation)AVAM · AVSSJ&F: 51.45 -> 52.94Improved
inject alignment only in blocks 3,4 instead of all blocks 1,2,3,4 (ablation)DDAVS · AVS-MS3J&F: 72.57 -> 76.01Improved
inject alignment in blocks 3,4 instead of block 1 only (ablation)DDAVS · AVS-MS3J&F: 68.37 -> 76.01Improved
inject in blocks 3,4 instead of block 3 only (ablation)DDAVS · AVS-MS3J&F: 74.27 -> 76.01Improved
use HTSAT audio encoder instead of VGGish (ablation)DDAVS (MiT-B5) · AVS-MS3J&F: 74.45 -> 76.01Improved
use MiT-B5 visual backbone instead of PVTv2-B5 (ablation)DDAVS (HTSAT) · AVS-MS3J&F: 74.69 -> 76.01Improved

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