Research map / Vision-language and multimodal / Cross-modal learning
Active Audio-Visual Separation of Dynamic Sound Sources
Sagnik Majumder, Kristen Grauman
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 transformer memory f^T (ablation) | Ours · Heard sounds (SI-SDR) | SI-SDR (dB): 2.32 -> 3.93 | Improved |
| add transformer memory f^T (ablation) | Ours · Unheard sounds (SI-SDR) | SI-SDR (dB): 2.02 -> 2.57 | Improved |
| add multi-step predictions (E=14 instead of 0) (ablation) | Ours · Heard sounds (SI-SDR) | SI-SDR (dB): 2.69 -> 3.93 | Improved |
| add multi-step predictions (E=14 instead of 0) (ablation) | Ours · Unheard sounds (SI-SDR) | SI-SDR (dB): 2.29 -> 2.57 | Improved |
| add visual input V_t (ablation) | Ours · Heard sounds (SI-SDR) | SI-SDR (dB): 3.61 -> 3.93 | Improved |
| add visual input V_t (ablation) | Ours · Unheard sounds (SI-SDR) | SI-SDR (dB): 2.40 -> 2.57 | Improved |
| use proposed active policy with transformer memory instead of Move2Hear | Move2Hear · Heard sounds (SI-SDR) | SI-SDR (dB): 2.31 -> 3.93 | Improved |
| use proposed agent instead of Move2Hear | Move2Hear · Unheard sounds (SI-SDR) | SI-SDR (dB): 2.06 -> 2.57 | Improved |
| use novelty-driven exploration policy instead of standing in place | Stand In-Place · Heard sounds (SI-SDR) | SI-SDR (dB): 2.49 -> 1.68 | Hurt |
| use Move2Hear instead of standing in place | Stand In-Place · Unheard sounds (SI-SDR) | SI-SDR (dB): 2.03 -> 2.06 | No clear effect |
| set multi-step parameter E to other non-zero values than 14 (ablation) | Ours · Unheard sounds (SI-SDR) | SI-SDR: drops up to 6.6% | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Move2Hear — Move2Hear: Active Audio-Visual Source Separation (ICCV 2021) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Move2Hear: Active Audio-Visual Source Separation (ICCV 2021 · similarity 0.94)
- Sample-efficient Audio-Visual Learning of Scene Acoustics (IJCV 2026 · similarity 0.90)
- Learning Audio-Visual Dynamics Using Scene Graphs for Audio Source Separation (NeurIPS 2022 · similarity 0.90)
- SoundSpaces: Audio-Visual Navigation in 3D Environments (ECCV 2020 · similarity 0.88)
- Continual Audio-Visual Sound Separation (NeurIPS 2024 · similarity 0.85)
- Towards Audio-Visual Navigation in Noisy Environments: A Large-Scale Benchmark Dataset and an Architecture Considering Multiple Sound-Sources (AAAI 2025 · similarity 0.85)
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.
