Research map / Video / Video generation
MoScale: Autoregressive Next-Scale Prediction for Human Motion Generation and Editing
Inwoo Hwang, Hojun Jang, Bing Zhou, Jian Wang, Young Min Kim, chuan guo
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 skeletal topology awareness to the tokenizer (ablation) | ScaleMoGen · SnapMoGen test | FID: 26.74 -> 16.35 | Improved |
| add skeletal topology awareness to the tokenizer (ablation) | ScaleMoGen · SnapMoGen test | CLIP Score: 0.630 -> 0.693 | Improved |
| reduce code size from 24 to 16 bits (ablation) | ScaleMoGen · SnapMoGen test | CLIP Score / FID: CLIP 0.693 -> 0.705; FID 16.35 -> 19.53 | Mixed |
| increase code size from 24 to 32 bits (ablation) | ScaleMoGen · SnapMoGen test | CLIP Score / FID: FID 16.35 -> 15.98; CLIP 0.693 -> 0.683 | Mixed |
| scale token map predictor from 125M to 2.2B (ablation) | ScaleMoGen · SnapMoGen test | CLIP Score: 0.693 -> 0.682 | Hurt |
| scale token map predictor from 125M to 362M (ablation) | ScaleMoGen · SnapMoGen test | FID: 16.35 -> 15.84 | No clear effect |
| use multi-scale skeletal-temporal bitwise VQ instead of categorical codebook (ablation) | ScaleMoGen · Motion reconstruction | FID: 4.83 -> 0.12 | Improved |
| remove skeletal topology from VQ (ablation) | ScaleMoGen · Motion reconstruction | MPJPE: 2.73 -> 6.66 | Improved |
| use ScaleMoGen next-scale prediction instead of MoMask | ScaleMoGen · HumanML3D test | FID: 0.045 -> 0.030 | Improved |
| use ScaleMoGen instead of MoMask++ | ScaleMoGen · SnapMoGen test | CLIP Score: 0.685 -> 0.693 | Improved |
| use ScaleMoGen instead of MoMask++ | ScaleMoGen · SnapMoGen test | FID: 15.06 -> 16.35 | Hurt |
| use ScaleMoGen editing instead of SALAD | ScaleMoGen · Motion editing user study | Overall quality: 3.23 -> 4.65 | Improved |
Closest papers by meaning
- MMM: Generative Masked Motion Model (CVPR 2024 · similarity 0.90)
- SnapMoGen: Human Motion Generation from Expressive Texts (NeurIPS 2025 · similarity 0.90)
- Scaling Large Motion Models with Million-Level Human Motions (ICML 2025 · similarity 0.89)
- SimMotionEdit: Text-Based Human Motion Editing with Motion Similarity Prediction (CVPR 2025 · similarity 0.89)
- ScaMo: Exploring the Scaling Law in Autoregressive Motion Generation Model (CVPR 2025 · similarity 0.89)
- Dynamic Motion Blending for Versatile Motion Editing (CVPR 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.
