Research map / Vision-language and multimodal / Cross-modal learning
Generate, Analyze, and Refine: Training-Free Sound Source Localization via MLLM Meta-Reasoning
Subin Park, Jung Uk Kim
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 |
|---|---|---|---|
| increase analysis iterations from N=1 to N=5 (ablation) | GAR · VGGSound-Duet | CIoU@0.3: 58.9 -> 77.6 | Improved |
| increase analysis iterations from N=1 to N=5 (ablation) | GAR · MUSIC-Duet | CIoU@0.3: 80.7 -> 82.7 | Improved |
| increase analysis iterations from N=1 to N=5 (ablation) | GAR · MUSIC-Solo | IoU@0.5: 96.3 -> 98.5 | Improved |
| increase analysis iterations from N=1 to N=3 (ablation) | GAR · VGGSound-Single | AP: 60.1 -> 60.2 | No clear effect |
| add Analysis and Refinement stages to Generation (ablation) | GAR · VGGSound-Duet | CIoU@0.3: 42.6 -> 59.5 | Improved |
| use Qwen2.5-Omni-7B instead of 3B backbone (ablation) | GAR · VGGSound-Duet | CIoU@0.3: 49.8 -> 59.5 | Improved |
| add audio class conditioning to refinement (Method 1 to Method 2) (ablation) | prompt variants · VGGSound-Single | IoU@0.5: 46.5 -> 59.0 | Improved |
| add audio class conditioning to refinement (Method 1 to Method 2) (ablation) | prompt variants · VGGSound-Duet | CIoU@0.3: 57.0 -> 23.0 | Hurt |
| add visual anchors to class-conditional refinement (Method 2 to Method 3) (ablation) | prompt variants · VGGSound-Duet | CIoU@0.3: 23.0 -> 60.4 | Improved |
| add meta-analysis info (consistency, role tags, voting) to anchor refinement (Method 3 to Ours N=1) (ablation) | prompt variants · MUSIC-Solo | AP: 81.6 -> 78.8 | Hurt |
| add meta-analysis info (consistency, role tags, voting) to anchor refinement (Method 3 to Ours N=1) (ablation) | prompt variants · MUSIC-Duet | CIoU@0.3: 76.9 -> 80.8 | Improved |
| raise audio confidence threshold from 0.5 to 0.75 (ablation) | GAR (N=5) · VGGSound-Single | AP: 60.1 -> 60.5 | No clear effect |
Closest papers by meaning
- Object-aware Sound Source Localization via Audio-Visual Scene Understanding (CVPR 2025 · similarity 0.91)
- Improving Sound Source Localization with Joint Slot Attention on Image and Audio (CVPR 2025 · similarity 0.89)
- Prompting Segmentation with Sound Is Generalizable Audio-Visual Source Localizer (AAAI 2024 · similarity 0.88)
- Hear, Localize, and Reason: Spatially Aware Scene Understanding for Audio-visual LLMs (NeurIPS 2026 · similarity 0.88)
- Seeing Sound, Hearing Sight: Uncovering Modality Bias and Conflict of AI models in Sound Localization (NeurIPS 2025 · similarity 0.88)
- T-VSL: Text-Guided Visual Sound Source Localization in Mixtures (CVPR 2024 · similarity 0.88)
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.
