Research map / Vision-language and multimodal / Cross-modal learning
Semantic Noise Reduction via Teacher-Guided Dual-Path Audio-Visual Representation Learning
Linge Wang, Yingying Chen, Bingke Zhu, Lu Zhou, Jinqiao Wang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use TG-DP dual-path training instead of single-pass CAV-MAE Sync | CAV-MAE Sync · AudioSet A->V zero-shot retrieval | R@1: 27.9 -> 37.1 | Improved |
| use TG-DP dual-path training instead of single-pass CAV-MAE Sync | CAV-MAE Sync · AudioSet V->A zero-shot retrieval | R@1: 35.2 -> 37.4 | Improved |
| use TG-DP instead of CAV-MAE Sync for unimodal probing | CAV-MAE Sync · AS20K audio-only | mAP: 29.3 -> 31.2 | Improved |
| use TG-DP instead of CAV-MAE Sync for unimodal probing | CAV-MAE Sync · AS20K vision-only | mAP: 14.3 -> 17.8 | Improved |
| use TG-DP instead of CAV-MAE Sync | CAV-MAE Sync · VGGSound attention probing | top-1 accuracy (%): 52.7 -> 52.7 | No clear effect |
| add a separate second forward pass for contrastive learning (ablation) | CAV-MAE Sync · VGGSound A->V retrieval | R@10: 58.1 -> 60.1 | Improved |
| add a separate second forward pass for contrastive learning (ablation) | CAV-MAE Sync · AS20K attention probing | mAP: 30.5 -> 30.4 | No clear effect |
| lower contrastive-branch mask ratio from 0.75 to 0.50 (ablation) | TG-DP · VGGSound V->A retrieval | R@1: 27.1 -> 31.3 | Improved |
| lower contrastive-branch mask ratio from 0.75 to 0.50 (ablation) | TG-DP · AS20K attention probing | mAP: 30.7 -> 32.0 | Improved |
| use no masking (0.00) instead of 0.50 in the contrastive branch (ablation) | TG-DP · AS20K attention probing | mAP: 32.0 -> 29.6 | Hurt |
| add teacher-student distillation (ablation) | TG-DP · AS20K attention probing | mAP: 30.5 -> 32.0 | Improved |
| add teacher-student distillation (ablation) | TG-DP · VGGSound V->A retrieval | R@1: 29.5 -> 31.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CAV-MAE Sync — CAV-MAE Sync: Improving Contrastive Audio-Visual Mask Autoencoders via Fine-Grained Alignment (CVPR 2025) · better on 7 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CAV-MAE Sync: Improving Contrastive Audio-Visual Mask Autoencoders via Fine-Grained Alignment (CVPR 2025 · similarity 0.90)
- From Vision to Audio and Beyond: A Unified Model for Audio-Visual Representation and Generation (ICML 2024 · similarity 0.88)
- Asymmetric Hierarchical Anchoring for Robust Audio–Visual Cross-Modal Generalization (NeurIPS 2026 · similarity 0.88)
- Seeing Speech and Sound: Distinguishing and Locating Audio Sources in Visual Scenes (CVPR 2025 · similarity 0.87)
- Dynamic Derivation and Elimination: Audio Visual Segmentation with Enhanced Audio Semantics (CVPR 2025 · similarity 0.87)
- Native Audio-Visual Alignment for Generation (NeurIPS 2026 · similarity 0.87)
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.
