Research map / Video / Video-language understanding
Understanding Co-speech Gestures in-the-wild
Sindhu B Hegde, K R Prajwal, Taein Kwon, Andrew Zisserman
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 gesture-word coupling loss to sequence contrastive loss (ablation) | JEGAL · AVS-Spot | accuracy: 20.83 -> 63.60 | Improved |
| add gesture-word coupling loss to sequence contrastive loss (ablation) | JEGAL · AVS-Ret speech-text to gesture | R@5: 12.20 -> 18.80 | Improved |
| add sequence contrastive loss to word coupling loss (ablation) | JEGAL · AVS-Asd (6 speakers) | accuracy: 14.8 -> 48.0 | Improved |
| add sequence contrastive loss to word coupling loss (ablation) | JEGAL · AVS-Ret speech-text to gesture | R@5: 8.50 -> 18.80 | Improved |
| concatenate rather than average speech and text embeddings (ablation) | JEGAL · AVS-Spot | accuracy: 56.04 -> 63.60 | Improved |
| concatenate rather than average speech and text embeddings (ablation) | JEGAL · AVS-Asd (6 speakers) | accuracy: 41.2 -> 48.0 | Improved |
| fuse speech and text before a single contrastive head instead of pairwise text loss (ablation) | JEGAL · AVS-Spot | accuracy: 34.31 -> 63.60 | Improved |
| fuse speech and text before a single contrastive head instead of pairwise audio loss (ablation) | JEGAL · AVS-Spot | accuracy: 23.67 -> 63.60 | Improved |
| add audio to text input at inference | JEGAL · AVS-Spot | accuracy: 61.00 -> 63.60 | Improved |
| add text to audio input at inference | JEGAL · AVS-Ret speech-text to gesture | R@5: 11.00 -> 18.80 | Improved |
| raise modality drop rate from 30% to 50% in training (text only inference) (ablation) | JEGAL · AVS-Spot | accuracy: 52.2 -> 61.0 | Improved |
| raise modality drop rate from 50% to 70% in training (ablation) | JEGAL · AVS-Spot (text+audio) | accuracy: 63.6 -> 62.6 | Hurt |
Closest papers by meaning
- Co$^{\mathbf{3}}$Gesture: Towards Coherent Concurrent Co-speech 3D Gesture Generation with Interactive Diffusion (ICLR 2025 · similarity 0.88)
- ConvoFusion: Multi-Modal Conversational Diffusion for Co-Speech Gesture Synthesis (CVPR 2024 · similarity 0.88)
- TANGO: Co-Speech Gesture Video Reenactment with Hierarchical Audio Motion Embedding and Diffusion Interpolation (ICLR 2025 · similarity 0.86)
- Retrieving Semantics from the Deep: an RAG Solution for Gesture Synthesis (CVPR 2025 · similarity 0.86)
- SocialGesture: Delving into Multi-person Gesture Understanding (CVPR 2025 · similarity 0.86)
- Co-Speech Gesture Video Generation with Implicit Motion-Audio Entanglement (CVPR 2025 · similarity 0.86)
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.
