Research map / Video / Video generation
The Quest for Generalizable Motion Generation: Data, Model, and Evaluation
Jing Lin, Ruisi Wang, Junzhe Lu, Ziqi Huang, Guorui Song, Ailing Zeng, Xian Liu, Chen Wei, Wanqi Yin, Qingping SUN, Zhongang Cai, Lei Yang and 1 more
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 adaptive gating between T2M and M2M branches instead of a single branch (ablation) | ViMoGen · MBench | Motion Generalizability: 0.54 (T2M only) / 0.59 (M2M only) -> 0.68 | Improved |
| add M2M branch conditioned on video motion tokens (M2M only vs video generation baseline) (ablation) | ViMoGen · MBench | Jitter Degree: 0.0193 -> 0.0145 | Improved |
| add T2M-only training on MoCap (vs adaptive) (ablation) | ViMoGen · MBench | Foot Sliding: 0.0113 (M2M) -> 0.0039 (T2M) | Improved |
| add synthetic ViGen-derived video data (14k clips) to training (ablation) | ViMoGen-light · MBench | Motion Generalizability: 0.50 -> 0.55 | Improved |
| add in-the-wild visual MoCap data to training (ablation) | ViMoGen-light · MBench | Foot Sliding: 0.0033 -> 0.0042 | Hurt |
| add other optical MoCap datasets to HumanML3D (ablation) | ViMoGen-light · MBench | Motion Generalizability: 0.44 -> 0.48 | Improved |
| use T5-XXL instead of CLIP text encoder (ablation) | ViMoGen · MBench | Motion Generalizability: 0.35 -> 0.44 | Improved |
| use MLLM text encoder instead of T5-XXL (ablation) | ViMoGen · MBench | Body Penetration: 1.05 -> 1.51 | Hurt |
| train on descriptive video-style text instead of motion-style text (ablation) | ViMoGen · MBench | Motion Generalizability: 0.40 -> 0.48 | Improved |
| test with video-style instead of motion-style text after video-style training (ablation) | ViMoGen · MBench | Motion Generalizability: 0.48 -> 0.44 | Hurt |
| use jointly trained shared-parameter branches instead of VLM-selected MotionLCM/DNO (ablation) | ViMoGen · MBench | Motion Generalizability: 0.63 -> 0.68 | Improved |
| use intermediate ViGen features as video tokens (ablation) | ViMoGen · MBench | convergence / performance | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- MotionCraft — MotionCraft: Physics-Based Zero-Shot Video Generation (NeurIPS 2024) · better on 1 of 2 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
- GENMO: A GENeralist Model for Human MOtion (ICCV 2025 · similarity 0.91)
- VMBench: A Benchmark for Perception-Aligned Video Motion Generation (ICCV 2025 · similarity 0.91)
- Human Motion Diffusion Model (ICLR 2023 · similarity 0.91)
- Rethinking Diffusion for Text-Driven Human Motion Generation: Redundant Representations, Evaluation, and Masked Autoregression (CVPR 2025 · similarity 0.90)
- VBench: Comprehensive Benchmark Suite for Video Generative Models (CVPR 2024 · similarity 0.89)
- SnapMoGen: Human Motion Generation from Expressive Texts (NeurIPS 2025 · similarity 0.89)
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.
