Research map / Video / Video generation
LiveGesture: Streamable Co-Speech Gesture Generation Model
Muhammad Usama Saleem, Mayur Jagdishbhai Patel, Ekkasit Pinyoanuntapong, Zhongxing Qin, Li Yang, Hongfei Xue, Ahmed Helmy, Chen Chen, Pu Wang
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 temporal attention in fusion transformer (ablation) | LiveGesture · BEAT2 | FGD: 15.52 -> 4.57 | Improved |
| add spatial attention in fusion transformer (ablation) | LiveGesture · BEAT2 | FGD: 6.64 -> 4.57 | Improved |
| add PILOR low-rank pre-infusion alignment (ablation) | LiveGesture · BEAT2 | FGD: 4.89 -> 4.57 | Improved |
| add uncertainty-guided refinement (UGR) (ablation) | LiveGesture · BEAT2 | BC: 0.723 -> 0.794 | Improved |
| add text cues (ablation) | LiveGesture · BEAT2 | FGD: 4.60 -> 4.57 | No clear effect |
| add xAR-Fuse fusion on top of region eXperts (ablation) | Local region eXperts · BEAT2 | FGD: 6.458 -> 4.57 | Improved |
| two-stage SVQ with only Quant+MLP trainable instead of 1-stage joint training (ablation) | SVQ tokenizer · BEAT2 | FGD: 4.691 -> 4.557 | Improved |
| increase codebook size from 512 to 2048 (ablation) | SVQ tokenizer · BEAT2 | FGD: 6.63 -> 4.51 | Improved |
| use causal cross-attention instead of additive causal self-attention (ablation) | HAR · BEAT2 | FGD: 4.63 -> 4.57 | Improved |
| use lightweight causal audio encoder instead of Magic-Codec (ablation) | audio encoder · BEAT2 | FGD: 4.51 -> 4.57 | No clear effect |
| set local expert loss weight 0.3 instead of 0.0 (ablation) | LiveGesture · BEAT2 | FGD: 4.89 -> 4.57 | Improved |
| set local expert loss weight 1.0 instead of 0.3 (ablation) | LiveGesture · BEAT2 | FGD: 4.57 -> 4.85 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- GestureLSM — GestureLSM: Latent Shortcut based Co-Speech Gesture Generation with Spatial-Temporal Modeling (ICCV 2025) · better on 1 of 3 shared results
- MambaTalk — MambaTalk: Efficient Holistic Gesture Synthesis with Selective State Space Models (NeurIPS 2024) · better on 3 of 3 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
- EMAGE: Towards Unified Holistic Co-Speech Gesture Generation via Expressive Masked Audio Gesture Modeling (CVPR 2024 · similarity 0.90)
- Co$^{\mathbf{3}}$Gesture: Towards Coherent Concurrent Co-speech 3D Gesture Generation with Interactive Diffusion (ICLR 2025 · similarity 0.89)
- Co-Speech Gesture Video Generation with Implicit Motion-Audio Entanglement (CVPR 2025 · similarity 0.89)
- Co-Speech Gesture Video Generation via Motion-Decoupled Diffusion Model (CVPR 2024 · similarity 0.88)
- TANGO: Co-Speech Gesture Video Reenactment with Hierarchical Audio Motion Embedding and Diffusion Interpolation (ICLR 2025 · similarity 0.88)
- ConvoFusion: Multi-Modal Conversational Diffusion for Co-Speech Gesture Synthesis (CVPR 2024 · similarity 0.88)
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.
