Research map / Video / Video segmentation and tracking
Telling Left From Right: Learning Spatial Correspondence of Sight and Sound
Karren Yang, Bryan Russell, Justin Salamon
Check 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 a mono-audio stream to the flipping-task model (ablation) | pretext classifier · YouTube-ASMR and FAIR-Play | accuracy: 80.1 -> 81.3; 93.6 -> 96.3 | Improved |
| replace ResNet-18 with MCx motion features (ablation) | pretext classifier · YouTube-ASMR and FAIR-Play | accuracy: 80.1 -> 80.4; 93.6 -> 71.3 | Mixed |
| add GCC-Phat features (ablation) | pretext classifier · YouTube-ASMR and FAIR-Play | accuracy: 80.1 -> 80.1; 93.6 -> 94.1 | No clear effect |
| use the proposed model instead of the strongest listed baseline | audio-visual face localization · YouTube-ASMR test | AP50: 27.9 -> 43.4 | Improved |
| initialize the localization model with flipping-task pretrained weights (ablation) | audio-visual face localization · YouTube-ASMR localization test | AP50: 23.9 -> 34.9 | Improved |
| use flipping-task visual features rather than ImageNet supervised features | audio upmixing · YouTube-ASMR | L1 error: 0.0858 -> 0.0853 | Improved |
| use flipping-task visual features rather than ImageNet supervised features | audio upmixing · FAIR-Play | L1 error: 0.403 -> 0.401 | Improved |
| combine flipping-task features with ImageNet features (ablation) | source separation · YouTube-ASMR | L1 error: 0.0885 -> 0.0863 | Improved |
| combine flipping-task features with ImageNet features (ablation) | source separation · FAIR-Play | L1 error: 0.362 -> 0.350 | Improved |
| initialize the DOA model with self-supervised 360-degree audio-visual pretrained weights | DOA estimation · Tau Spatial Sound | mean error: 20.5 -> 17.5 | Improved |
| use self-supervised embeddings rather than random embeddings for one-shot SVM | DOA estimation · Tau Spatial Sound one-shot | mean error: 78.6 -> 29.5 | Improved |
Closest papers by meaning
- Learning Spatial Features from Audio-Visual Correspondence in Egocentric Videos (CVPR 2024 · similarity 0.91)
- Learning Representations from Audio-Visual Spatial Alignment (NeurIPS 2020 · similarity 0.90)
- Hear, Localize, and Reason: Spatially Aware Scene Understanding for Audio-visual LLMs (NeurIPS 2026 · similarity 0.89)
- The World is Not Mono: Enabling Spatial Understanding in Large Audio-Language Models (NeurIPS 2026 · similarity 0.88)
- OWL : Geometry-Aware Spatial Reasoning for Audio Large Language Models (ICLR 2026 · similarity 0.87)
- SAVVY: Spatial Awareness via Audio-Visual LLMs through Seeing and Hearing (NeurIPS 2025 · 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.
