Research map / Vision-language and multimodal / Cross-modal learning
Learning to Localize Sound Sources in Visual Scenes: Analysis and Applications
Arda Senocak, Tae-Hyun Oh, Junsik Kim, Ming–Hsuan Yang, In So Kweon
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 |
|---|---|---|---|
| add 2.5k supervised samples to unsupervised 10k training (ablation) | sound localization network · Flickr-SoundNet (softmax) | cIoU: 43.6 -> 82.8 | Improved |
| train on 144k unsupervised pairs instead of 10k (ablation) | unsupervised network · Flickr-SoundNet (softmax) | cIoU: 43.6 -> 66.0 | Improved |
| add unsupervised 10k loss to supervised 2.5k (ablation) | supervised network · Flickr-SoundNet (softmax) | cIoU: 80.4 -> 82.8 | Improved |
| add unsupervised 10k loss to supervised 2.5k (ablation) | supervised network · Flickr-SoundNet (ReLU+softmax) | cIoU: 82.0 -> 84.0 | Improved |
| add 144k unsupervised pairs instead of 10k to semi-supervised model (ablation) | semi-supervised network · Flickr-SoundNet (ReLU+softmax) | cIoU: 84.0 -> 84.4 | No clear effect |
| increase supervised samples from 0.10k to 1.00k with 10k unsupervised (ablation) | semi-supervised network · Flickr-SoundNet (ReLU+softmax) | cIoU: 77.2 -> 82.4 | Improved |
| increase supervised samples beyond 1.00k to 2.50k (ablation) | semi-supervised network · Flickr-SoundNet (ReLU+softmax) | cIoU: 82.4 -> 84.0 | No clear effect |
| add ReLU to attention (suppress negative correlation) for unsupervised 144k (ablation) | unsupervised network · Flickr-SoundNet | cIoU: 66.0 -> 52.4 | Hurt |
| add ReLU to attention for supervised 2.5k (ablation) | supervised network · Flickr-SoundNet | cIoU: 80.4 -> 82.0 | Improved |
| add ReLU to attention (ablation) | localization network · ambient sounds (qualitative) | response to ambient sound | Improved |
| set loss weight Unsup.=0.5, Sup.=0.5 versus Unsup.=0.1, Sup.=0.9 (ablation) | semi-supervised network · Unsup. 144k + Sup. 0.5k | cIoU: 82.8 -> 85.2 | Improved |
| raise unsupervised loss weight to 0.9 (ablation) | semi-supervised network · Unsup. 144k + Sup. 0.5k | cIoU: 85.2 -> 83.2 | Hurt |
Closest papers by meaning
- Learning to Localize Sound Source in Visual Scenes (CVPR 2018 · similarity 0.97)
- Localizing Visual Sounds the Easy Way (ECCV 2022 · similarity 0.93)
- A Closer Look at Weakly-Supervised Audio-Visual Source Localization (NeurIPS 2022 · similarity 0.93)
- Sound to Visual Scene Generation by Audio-to-Visual Latent Alignment (CVPR 2023 · similarity 0.92)
- 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 0.90)
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.
