Research map / Video / Video generation
Next-Scale Autoregressive Models for Text-to-Motion Generation
Zhiwei Zheng, Shibo Jin, Lingjie Liu, Mingmin Zhao
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-scale hierarchical refinement (ablation) | MoScale · HumanML3D test | Top-1: 0.481 -> 0.534 | Improved |
| add cross-scale hierarchical refinement (ablation) | MoScale · HumanML3D test | FID: 0.176 -> 0.090 | Improved |
| add in-scale temporal refinement (ablation) | MoScale base · HumanML3D test | Top-1: 0.481 -> 0.497 | Improved |
| add temporal refinement on top of hierarchical refinement (ablation) | MoScale · HumanML3D test | FID: 0.090 -> 0.046 | Improved |
| add temporal refinement on top of hierarchical refinement (ablation) | MoScale · HumanML3D test | Top-1: 0.534 -> 0.540 | Improved |
| add temporal refinement only (ablation) | MoScale base · HumanML3D high-complexity descriptions | VLM text-alignment score: 1.89 -> 1.92 | No clear effect |
| use ordered coarse-to-fine generation instead of undirected mask-and-repredict (ablation) | MoScale · HumanML3D test | Top-1: 0.507 -> 0.540 | Improved |
| raise max corruption rate from 0.2 to 0.6 (ablation) | hierarchical refinement · HumanML3D test | FID: 0.096 -> 0.046 | Improved |
| raise max corruption rate from 0.6 to 0.8 (ablation) | hierarchical refinement · HumanML3D test | FID: 0.046 -> 0.074 | Hurt |
| increase transformer depth from 4 to 16 layers (ablation) | MoScale · HumanML3D test | FID: 0.096 -> 0.046 | Improved |
| increase refinement iterations beyond (1,2,5,10) (ablation) | MoScale · HumanML3D test | Top-1: 0.540 -> 0.539 | No clear effect |
| increase refinement iterations from (1,1,1,1) to (1,2,5,10) (ablation) | MoScale · HumanML3D test | Top-1: 0.535 -> 0.540 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ParCo — ParCo: Part-Coordinating Text-to-Motion Synthesis (ECCV 2024) · better on 4 of 4 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
- MoScale: Autoregressive Next-Scale Prediction for Human Motion Generation and Editing (ECCV 2026 · similarity 0.92)
- ScaMo: Exploring the Scaling Law in Autoregressive Motion Generation Model (CVPR 2025 · similarity 0.90)
- BAMM: Bidirectional Autoregressive Motion Model (ECCV 2024 · similarity 0.90)
- DartControl: A Diffusion-Based Autoregressive Motion Model for Real-Time Text-Driven Motion Control (ICLR 2025 · similarity 0.88)
- AMD: Autoregressive Motion Diffusion (AAAI 2024 · similarity 0.88)
- MMM: Generative Masked Motion Model (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.
