Research map / Vision-language and multimodal / Cross-modal learning
Sound to Visual Scene Generation by Audio-to-Visual Latent Alignment
Kim Sung-Bin, Arda Senocak, Hyunwoo Ha, Andrew Owens, Tae-Hyun Oh
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 the image encoder input with an audio encoder while retaining the same generator | Sound2Scene · VGGSound (50 classes) | R@1: 30.06 -> 40.71 | Improved |
| Replace the image encoder input with an audio encoder while retaining the same generator | Sound2Scene · VGGSound (50 classes) | FID: 16.11 -> 17.97 | Hurt |
| Replace the image encoder input with an audio encoder while retaining the same generator | Sound2Scene · VEGAS | R@1: 46.60 -> 57.44 | Improved |
| Generate images rather than retrieve them from the training-image database | Sound2Scene · VGGSound (50 classes) | R@1: 51.28 -> 40.71 | Hurt |
| Generate images rather than retrieve them from the training-image database | Sound2Scene · VGGSound (50 classes) | R@5: 80.37 -> 77.36 | Hurt |
| Use symmetric InfoNCE rather than L2 feature matching (ablation) | Sound2Scene · VGGSound (50 classes) | R@1: 18.21 -> 40.71 | Improved |
| Use symmetric InfoNCE rather than one-way InfoNCE (ablation) | Sound2Scene · VGGSound (50 classes) | R@1: 31.63 -> 40.71 | Improved |
| Add sound-source-localization frame selection (ablation) | Sound2Scene · VGGSound (50 classes) | R@1: 37.20 -> 40.71 | Improved |
| Increase audio duration from 1 second to 5 seconds (ablation) | Sound2Scene · VGGSound (50 classes) | R@1: 35.85 -> 38.24 | Improved |
| Increase audio duration from 5 seconds to 10 seconds (ablation) | Sound2Scene · VGGSound (50 classes) | R@1: 38.24 -> 40.71 | Improved |
| Increase audio duration from 1 second to 10 seconds (ablation) | Sound2Scene · VGGSound (50 classes) | FID: 19.05 -> 17.97 | Improved |
| Increase audio duration from 1 second to 10 seconds (ablation) | Sound2Scene · VGGSound (50 classes) | IS: 17.87 -> 19.46 | Improved |
Closest papers by meaning
- Sounding that Object: Interactive Object-Aware Image to Audio Generation (ICML 2025 · similarity 0.91)
- Object-aware Sound Source Localization via Audio-Visual Scene Understanding (CVPR 2025 · similarity 0.91)
- Robust Audio-Visual Segmentation via Audio-Guided Visual Convergent Alignment (CVPR 2025 · similarity 0.90)
- Seeing and Hearing: Open-domain Visual-Audio Generation with Diffusion Latent Aligners (CVPR 2024 · similarity 0.90)
- From Vision to Audio and Beyond: A Unified Model for Audio-Visual Representation and Generation (ICML 2024 · similarity 0.90)
- Seeing Speech and Sound: Distinguishing and Locating Audio Sources in Visual Scenes (CVPR 2025 · similarity 0.89)
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.
