Research map / Vision-language and multimodal / Human and embodied modeling
UniMotion: A Unified Framework for Motion-Text-Vision Understanding and Generation
Ziyi Wang, Xinshun Wang, Shuang Chen, Yang Cong, Mengyuan Liu
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 Dual-Posterior KL Alignment (DPA) (ablation) | UniMotion · HumanML3D text-to-motion | R-Precision Top-3: 0.818 -> 0.841 | Improved |
| add DPA (ablation) | UniMotion · MotionFix motion editing | R@3: 80.35 -> 84.94 | Improved |
| add DPA (ablation) | UniMotion · Vision-to-motion | MPJPE: 83.1 -> 75.0 | Improved |
| add Latent Reconstruction Alignment (LRA) pre-training (ablation) | UniMotion · HumanML3D text-to-motion | R-Precision Top-3: 0.801 -> 0.841 | Improved |
| add LRA pre-training (ablation) | UniMotion · Motion prediction | ADE: 3.777 -> 3.172 | Improved |
| add LRA pre-training (ablation) | UniMotion · Vision-to-motion | MPJPE: 84.3 -> 75.0 | Improved |
| use hybrid attention (full within motion spans) instead of global causal only (ablation) | UniMotion · MotionFix motion editing | R@3: 79.6 -> 84.94 | Improved |
| use modality-routed LoRA instead of shared LoRA (ablation) | UniMotion · Vision-to-motion | MPJPE: 90.4 -> 75.0 | Improved |
| add LoRA adaptation to frozen LLM (ablation) | UniMotion · HumanML3D text-to-motion | R-Precision Top-3: 0.752 -> 0.818 | Improved |
| replace VQ-VAE with continuous CMA-VAE (ablation) | UniMotion Gen-Branch embedder · HumanML3D text-to-motion | R-Precision Top-3: 0.752 -> 0.824 | Improved |
| add the second embedder branch (dual-path) to generation branch only (ablation) | UniMotion with CMA-VAE · Motion prediction | ADE: 4.862 -> 3.172 | Improved |
| add semantic branch to generation-only: M2T trade-off between branches (ablation) | UniMotion with CMA-VAE · HumanML3D motion-to-text vs text-to-motion | BertScore / R@3: 40.1 vs 36.3; 0.824 vs 0.798 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MotionGPT — MotionGPT: Human Motion as a Foreign Language (NeurIPS 2023) · better on 4 of 4 shared results
- MG-MotionLLM — MG-MotionLLM: A Unified Framework for Motion Comprehension and Generation across Multiple Granularities (CVPR 2025) · 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
- Harnessing Frozen Unimodal Encoders for Flexible Multimodal Alignment (CVPR 2025 · similarity 0.89)
- HIMMEL: Hierarchical Interleaved Multi-stream Motion Encoding for Long Video Understanding (NeurIPS 2026 · similarity 0.88)
- Tri-Modal Motion Retrieval by Learning a Joint Embedding Space (CVPR 2024 · similarity 0.87)
- HMVLM:Human Motion-Vision-Language Model via MoE LoRA (NeurIPS 2025 · similarity 0.87)
- LaMP: Language-Motion Pretraining for Motion Generation, Retrieval, and Captioning (ICLR 2025 · similarity 0.87)
- Enhanced Motion-Text Alignment for Image-to-Video Transfer Learning (CVPR 2024 · similarity 0.87)
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.
