Research map / Vision-language and multimodal / Cross-modal learning
MAAS: Multi-Modal Assignation for Active Speaker Detection
Juan Léon Alcázar, Fabian Caba, Ali K. Thabet, Bernard Ghanem
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 static graph stream to dynamic stream (ablation) | MAAS-TAN · AVA-ActiveSpeaker val | mAP: 66.5 -> 88.8 | Improved |
| add dynamic graph stream to static stream (ablation) | MAAS-TAN · AVA-ActiveSpeaker val | mAP: 87.9 -> 88.8 | Improved |
| increase GCN depth from 1 to 4 layers (ablation) | MAAS-TAN · AVA-ActiveSpeaker val | mAP: 88.0 -> 88.8 | Improved |
| increase GCN depth from 4 to 5 layers (ablation) | MAAS-TAN · AVA-ActiveSpeaker val | mAP: 88.8 -> 87.5 | Hurt |
| increase filters from 64 to 256 (ablation) | MAAS-TAN · AVA-ActiveSpeaker val | mAP: 88.8 -> 88.1 | Hurt |
| increase filters from 32 to 64 (ablation) | MAAS-TAN · AVA-ActiveSpeaker val | mAP: 88.5 -> 88.8 | Improved |
| use K=3 instead of K=2 neighbors (ablation) | MAAS-TAN · AVA-ActiveSpeaker val | mAP: 88.5 -> 88.8 | Improved |
| use K=3 instead of K=5 neighbors (ablation) | MAAS-TAN · AVA-ActiveSpeaker val | mAP: 88.4 -> 88.8 | Improved |
| extend from 1 LAN to 13 LANs (temporal graph) with 4 video nodes (ablation) | MAAS · AVA-ActiveSpeaker val | mAP: 85.1 -> 88.8 | Improved |
| increase LANs from 13 to 15 (ablation) | MAAS-TAN · AVA-ActiveSpeaker val | mAP: 88.8 -> 88.5 | Hurt |
| increase video nodes per LAN from 1 to 4 with 1 LAN (ablation) | MAAS · AVA-ActiveSpeaker val | mAP: 80.2 -> 85.1 | Improved |
| add supervision on audio nodes (ablation) | MAAS-TAN · AVA-ActiveSpeaker val | mAP: 88.5 -> 88.8 | Improved |
Closest papers by meaning
- A Light Weight Model for Active Speaker Detection (CVPR 2023 · similarity 0.91)
- Active Speakers in Context (CVPR 2020 · similarity 0.90)
- Cross-Modal Supervision for Learning Active Speaker Detection in Video (ECCV 2016 · similarity 0.89)
- LoCoNet: Long-Short Context Network for Active Speaker Detection (CVPR 2024 · similarity 0.87)
- Seeing Speech and Sound: Distinguishing and Locating Audio Sources in Visual Scenes (CVPR 2025 · similarity 0.83)
- Who Says What: Symbolic Trimodal Binding Mechanisms in Audio-Visual LLMs (NeurIPS 2026 · similarity 0.83)
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.
