Research map / Vision-language and multimodal / Cross-modal learning
How Far Can We Go With Synthetic Data for Audio-Visual Sound Source Localization?
Arda Senocak, Sooyoung Park, Tae-Hyun Oh, Joon Son Chung
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 |
|---|---|---|---|
| replace real VGGSound training data with fully synthetic data at the same scale (ablation) | ACL-SSL · Single Source Localization (1x) | cIoU: 48.03 -> 47.97 | No clear effect |
| replace real images with synthetic images, keep real audio (ablation) | ACL-SSL · Single Source Localization (1x) | cIoU: +7.10 cIoU, +4.94 cIoU Adap. | Improved |
| replace real images with synthetic images, keep real audio (ablation) | ACL-SSL · Interactive Localization (1x) | IIoU: +10.40 IIoU, +10.91 IIoU Adap. | Improved |
| replace real images with synthetic images, keep real audio (ablation) | ACL-SSL · Audio-Visual Segmentation (1x) | mIoU: +4.61 mIoU, +3.50 mIoU Adap. | Improved |
| replace real audio with synthetic audio, keep real images (ablation) | ACL-SSL · Single Source Localization (1x) | cIoU: 48.03 -> 46.38 | Hurt |
| add 10K real audio to synthetic images and synthetic audio (ablation) | ACL-SSL · Single Source Localization (1x) | cIoU: +4.20 cIoU, +2.53 cIoU Adap. vs Original | Improved |
| scale fully synthetic data to 2x (ablation) | ACL-SSL · Single Source Localization | cIoU: 47.11 (2x) vs 48.03 original | No clear effect |
| scale fully synthetic data to 3x (ablation) | ACL-SSL · Single Source Localization | cIoU: +0.87 cIoU, -0.54 cIoU Adap. vs Original | Improved |
| scale synthetic images with real audio from 1x to 3x (ablation) | ACL-SSL · Single Source Localization | cIoU: 55.13 -> 54.68 (2x) -> 55.73 (3x) | Improved |
| add synthetic image-real audio pairs on top of full real VGGSound (3x) | ACL-SSL · Single Source Localization | cIoU: +8.62 cIoU, +7.00 cIoU Adap. vs Original | Improved |
| add synthetic image-real audio pairs on top of full real VGGSound (3x) | ACL-SSL · Interactive Localization | IIoU: +12.16 IIoU, +13.35 IIoU Adap. | Improved |
| add synthetic image-real audio pairs on top of full real VGGSound (3x) | ACL-SSL · Audio-Visual Segmentation | mIoU: +5.58 mIoU, +4.22 mIoU Adap. | Improved |
Closest papers by meaning
- Object-aware Sound Source Localization via Audio-Visual Scene Understanding (CVPR 2025 · similarity 0.87)
- T-VSL: Text-Guided Visual Sound Source Localization in Mixtures (CVPR 2024 · similarity 0.87)
- Synthio: Augmenting Small-Scale Audio Classification Datasets with Synthetic Data (ICLR 2025 · similarity 0.87)
- V2A-Mapper: A Lightweight Solution for Vision-to-Audio Generation by Connecting Foundation Models (AAAI 2024 · similarity 0.87)
- Improving Sound Source Localization with Joint Slot Attention on Image and Audio (CVPR 2025 · similarity 0.86)
- Contrastive Learning from Synthetic Audio Doppelgängers (ICLR 2025 · similarity 0.86)
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.
