Research map / Video / Video generation
Role-Aware Interaction Generation from Textual Description
Mikihiro Tanaka, Kent Fujiwara
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 cross attention between the two human motion transformers (ablation) | two-Transformer baseline · NTU-RGB+D 120 interactions | Mutual Consistency: 46.3 -> 98.3 | Improved |
| add cross attention between the two human motion transformers (ablation) | two-Transformer baseline · NTU-RGB+D 120 interactions | FID: 21.1 -> 13.1 | Improved |
| initialize with pretrained single-person motion model (ablation) | Ours · NTU-RGB+D 120 interactions | Accuracy: 76.1 -> 79.9 | Improved |
| initialize with pretrained single-person motion model (ablation) | Ours · NTU-RGB+D 120 interactions | FID: 13.1 -> 12.8 | No clear effect |
| use role-aware two-transformer model instead of single transformer (ablation) | interaction generation · NTU-RGB+D 120 interactions | FID: 12.3 -> 13.1 | No clear effect |
| use pseudo-label role assignment from PIT model rather than label-based PIT generation (ablation) | role-aware model · NTU-RGB+D 120 interactions | Accuracy: 70.5 -> 76.1 | Improved |
| use per-motion transformers with cross attention instead of compressing both motions into one latent (MUGL, DSAG) | interaction generation · NTU-RGB+D 120 interactions | Mutual Consistency: 88.6 -> 98.3 (vs DSAG) | Improved |
| use noise step around t=800 for actor/receiver separation instead of early steps (ablation) | PIT role separation · NTU-RGB+D 120 asymmetric interactions | separation accuracy | Improved |
| use ensemble majority vote during role identification (ablation) | PIT role separation · NTU-RGB+D 120 asymmetric interactions | separation accuracy | Improved |
| use Mutual Consistency in addition to FID and accuracy | evaluation · NTU-RGB+D 120 interactions | metric sensitivity: Baseline (Two Trans.) accuracy 73.2 but Mutual Consistency 46.3 | Improved |
Closest papers by meaning
- InterDreamer: Zero-Shot Text to 3D Dynamic Human-Object Interaction (NeurIPS 2024 · similarity 0.87)
- A Unified Framework for Motion Reasoning and Generation in Human Interaction (ICCV 2025 · similarity 0.87)
- Text2Interact: High-Fidelity and Diverse Text-to-Two-Person Interaction Generation (ICLR 2026 · similarity 0.87)
- Learning to Generate Human-Human-Object Interactions from Textual Descriptions (NeurIPS 2025 · similarity 0.86)
- ReGenNet: Towards Human Action-Reaction Synthesis (CVPR 2024 · similarity 0.85)
- InterControl: Zero-shot Human Interaction Generation by Controlling Every Joint (NeurIPS 2024 · similarity 0.85)
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.
