Research map / Video / Video generation
Unifying Precise Keyframes and Semantic Control via Multi-level Diffusion
Linjun Wu, Jiejia Yu, Leyang Jin, He Wang, Bowen Zheng, Xu Yang, Hao Jiang, Fei Xia, Fei Ling, Jun Deng, Xiaogang Jin
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 |
|---|---|---|---|
| combine global and local keyframe guidance (ablation) | multi-level diffusion model · HumanML3D keyframe in-betweening | Keyframe Error (cm): 3.709 -> 2.933 (vs LocG); 9.464 -> 2.933 (vs GloG) | Improved |
| add global keyframe guidance to local guidance (ablation) | multi-level diffusion model · HumanML3D keyframe in-betweening | SS Similarity: 0.751 -> 0.806 | Improved |
| add trajectory refinement (ablation) | LocG + GloG · HumanML3D keyframe in-betweening | Keyframe Error (cm): 2.933 -> 0.667 | Improved |
| add trajectory refinement (ablation) | LocG + GloG · HumanML3D keyframe in-betweening | FID: 0.037 -> 0.032 | Improved |
| add pose refinement (diffusion imputation) (ablation) | LocG + GloG + TraR · HumanML3D keyframe in-betweening | Keyframe Error (cm): 0.667 -> 0.000 | Improved |
| add pose refinement (ablation) | LocG + GloG + TraR · HumanML3D keyframe in-betweening | FID: 0.032 -> 0.023 | Improved |
| add pose refinement (ablation) | LocG + GloG + TraR · HumanML3D keyframe in-betweening | SS Similarity: 0.807 -> 0.804 | No clear effect |
| use proportional-to-root-speed trajectory correction instead of uniform velocity adjustment (ablation) | trajectory refinement · HumanML3D keyframe in-betweening | foot skating / drift | Improved |
| increase number of intermediate keyframes from K=0 to K=20 (ablation) | multi-level diffusion model · HumanML3D keyframe sparsity | SS Similarity / FID: SS 0.719 -> 0.833; FID 0.159 -> 0.004 | Improved |
| increase number of intermediate keyframes (ablation) | multi-level diffusion model · HumanML3D keyframe sparsity | R-Precision (Top 3): 0.796 -> 0.802 | No clear effect |
| use invert-and-sample editing with strict keyframe constraints instead of DNO latent optimization | motion editing · HumanML3D pose editing | FID: 1.798 -> 0.284 | Improved |
Closest papers by meaning
- Enhanced Fine-Grained Motion Diffusion for Text-Driven Human Motion Synthesis (AAAI 2024 · similarity 0.93)
- Dynamic Motion Blending for Versatile Motion Editing (CVPR 2025 · similarity 0.89)
- DartControl: A Diffusion-Based Autoregressive Motion Model for Real-Time Text-Driven Motion Control (ICLR 2025 · similarity 0.88)
- Human Motion Diffusion as a Generative Prior (ICLR 2024 · similarity 0.88)
- Deep Compositional Phase Diffusion for Long Motion Sequence Generation (NeurIPS 2025 · similarity 0.87)
- 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.
