Research map / Detection and segmentation / CNN features
Active Speakers in Context
Juan Leon Alcazar, Fabian Caba, Long Mai, Federico Perazzi, Joon-Young Lee, Pablo Arbelaez, Bernard Ghanem
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 long-term multi-speaker context to a single-speaker short-term model (ablation) | Active Speakers Context · AVA-ActiveSpeaker validation | mAP: 79.5 -> 84.4 | Improved |
| add pairwise refinement (ablation) | context model without refinement · AVA-ActiveSpeaker validation | mAP: 84.4 -> 85.2 | Improved |
| add temporal refinement to pairwise-refined context (ablation) | Active Speakers Context · AVA-ActiveSpeaker validation | mAP: 85.2 -> 87.1 | Improved |
| replace temporal refinement with a parameter-matched two-layer MLP (ablation) | context with pairwise refinement · AVA-ActiveSpeaker validation | mAP: 85.3 vs 87.1 | Hurt |
| replace temporal refinement with a 0.5s moving average (ablation) | Active Speakers Context · AVA-ActiveSpeaker validation | mAP: +0.02% | No clear effect |
| replace temporal refinement with a 2.25s moving average (ablation) | Active Speakers Context · AVA-ActiveSpeaker validation | mAP: -11.64% | Hurt |
| increase temporal support from L=1 to L=11 while using three speakers (ablation) | Active Speakers Context · AVA-ActiveSpeaker validation | mAP: 82.9 -> 87.1 | Improved |
| increase context speakers from S=1 to S=3 at L=11 (ablation) | Active Speakers Context · AVA-ActiveSpeaker validation | mAP: 85.6 -> 87.1 | Improved |
| shuffle context clips out of temporal order (ablation) | Active Speakers Context · AVA-ActiveSpeaker validation | mAP: 87.1 -> 77.8 | Hurt |
| replace surrounding speakers with speakers sampled at another video time (ablation) | Active Speakers Context · AVA-ActiveSpeaker validation | mAP: 87.1 -> 84.5 | Hurt |
Closest papers by meaning
- A Light Weight Model for Active Speaker Detection (CVPR 2023 · similarity 0.91)
- How To Design a Three-Stage Architecture for Audio-Visual Active Speaker Detection in the Wild (ICCV 2021 · similarity 0.91)
- LoCoNet: Long-Short Context Network for Active Speaker Detection (CVPR 2024 · similarity 0.90)
- MAAS: Multi-Modal Assignation for Active Speaker Detection (ICCV 2021 · similarity 0.90)
- End-to-End Active Speaker Detection (ECCV 2022 · similarity 0.90)
- Who Says What: Symbolic Trimodal Binding Mechanisms in Audio-Visual LLMs (NeurIPS 2026 · similarity 0.82)
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.
