Research map / Graph learning / Knowledge and temporal graphs
Learning Long-Term Spatial-Temporal Graphs for Active Speaker Detection
Kyle Min, Sourya Roy, Subarna Tripathi, Tanaya Guha, Somdeb Majumdar
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 audio negative-sampling augmentation (ablation) | feature encoder · AVA-ActiveSpeaker val | mAP: 80.2 -> 82.6 | Improved |
| add TSM to the visual encoder (ablation) | SPELL · AVA-ActiveSpeaker val | mAP: 82.6 -> 88.0 | Improved |
| add context-reasoning GNN with undirected graph (ablation) | SPELL · AVA-ActiveSpeaker val | mAP: 88.0 -> 92.4 | Improved |
| add forward/backward graphs (Bi-dir) (ablation) | SPELL · AVA-ActiveSpeaker val | mAP: 92.4 -> 93.9 | Improved |
| add spatial features of face location (ablation) | SPELL · AVA-ActiveSpeaker val | mAP: 93.9 -> 94.2 | Improved |
| use only visual features (remove audio) (ablation) | SPELL · AVA-ActiveSpeaker val | mAP: 94.2 -> 84.9 | Hurt |
| use only audio features (remove visual) (ablation) | SPELL · AVA-ActiveSpeaker val | mAP: 94.2 -> 55.6 | Hurt |
| increase filter dimension from 64 to 128 (ablation) | SPELL · AVA-ActiveSpeaker val | mAP: 94.2 -> 94.1 | No clear effect |
| increase filter dimension from 64 to 256 (ablation) | SPELL · AVA-ActiveSpeaker val | mAP: 94.2 -> 94.1 | No clear effect |
| decrease filter dimension from 64 to 16 (ablation) | SPELL · AVA-ActiveSpeaker val | mAP: 94.2 -> 93.5 | Hurt |
| connect too distant face-crops (larger tau) (ablation) | SPELL · AVA-ActiveSpeaker val | mAP | Hurt |
| use a larger number of graph nodes (ablation) | SPELL · AVA-ActiveSpeaker val | mAP | No clear effect |
Closest papers by meaning
- End-to-End Active Speaker Detection (ECCV 2022 · similarity 0.91)
- LoCoNet: Long-Short Context Network for Active Speaker Detection (CVPR 2024 · similarity 0.89)
- Who Says What: Symbolic Trimodal Binding Mechanisms in Audio-Visual LLMs (NeurIPS 2026 · similarity 0.83)
- LiveCC: Learning Video LLM with Streaming Speech Transcription at Scale (CVPR 2025 · similarity 0.82)
- Heuristic-free Knowledge Distillation for Streaming ASR via Multi-modal Training (AAAI 2025 · similarity 0.82)
- ReWind: Understanding Long Videos with Instructed Learnable Memory (CVPR 2025 · 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.
