Research map / Video / Video generation
ReMoDiffuse: Retrieval-Augmented Motion Diffusion Model
Mingyuan Zhang, Xinying Guo, Liang Pan, Zhongang Cai, Fangzhou Hong, Huirong Li, Lei Yang, Ziwei 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 hybrid retrieval and semantics-modulated attention to the diffusion model (ablation) | MotionDiffuse-based baseline · KIT-ML test | FID: 0.245 -> 0.155 | Improved |
| use both semantic and kinematic similarity in retrieval instead of semantic only (ablation) | ReMoDiffuse retrieval · KIT-ML test | FID: 0.307 -> 0.155 | Improved |
| use text features in the attention over retrieved samples (ablation) | ReMoDiffuse · KIT-ML test | FID: 0.192 -> 0.155 | Improved |
| retrieve semantics only (lambda=0 kinematic weight) instead of balancing with length | hybrid retrieval · KIT-ML and HumanML3D | retrieval FID | Hurt |
| use 2 retrieved samples instead of 1 or 3 (ablation) | ReMoDiffuse · KIT-ML test | FID: 0.186 (1) / 0.155 (2) / 0.217 (3) | Mixed |
| add retrieval to the baseline for rare prompts (ablation) | ReMoDiffuse · KIT-ML test | tail 5% MM Dist: 6.173 -> 5.439 | Improved |
| apply ReMoDiffuse vs MotionDiffuse on HumanML3D | text-to-motion diffusion · HumanML3D test | FID: 0.630 -> 0.103 | Improved |
| use retrieval-augmented model on diversity metrics | ReMoDiffuse vs MDM/MotionDiffuse · HumanML3D test | Diversity: 9.410 (MotionDiffuse) -> 9.018 | Hurt |
Later papers that compared with this method
15 later accepted papers report ReMoDiffuse as a baseline in their results tables; 7 of them report a better number on most of the results they share with it.
- DrawMotion: Generating 3D Human Motions by Freehand Drawing (TPAMI 2026 · better on 2 of 4 shared results)
- K-Mask: Kinematic-Aware Masked Modeling for Controllable Text-to-Motion Synthesis (ECCV 2026 · better on 2 of 2 shared results)
- ReMoMask: Retrieval-Augmented Masked Motion Generation (ECCV 2026 · better on 2 of 2 shared results)
- StreamTalk: Streaming Co-Speech Gesture Generation with Key-Pose Anchoring (ECCV 2026 · better on 2 of 2 shared results)
- Reconstruction-Anchored Diffusion Model for Text-to-Motion Generation (ECCV 2026 · better on 0 of 1 shared result)
- StickMotion: Generating 3D Human Motions by Drawing a Stickman (CVPR 2025 · better on 13 of 14 shared results)
- Fg-T2M++: LLMs-Augmented Fine-Grained Text Driven Human Motion Generation (IJCV 2025 · better on 7 of 7 shared results)
- Less is More: Improving Motion Diffusion Models with Sparse Keyframes (ICCV 2025 · better on 0 of 2 shared results)
7 more not listed.
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
- ReMoGPT: Part-Level Retrieval-Augmented Motion-Language Models (AAAI 2025 · similarity 0.92)
- MotionMix: Weakly-Supervised Diffusion for Controllable Motion Generation (AAAI 2024 · similarity 0.91)
- Towards Detailed Text-to-Motion Synthesis via Basic-to-Advanced Hierarchical Diffusion Model (AAAI 2024 · similarity 0.90)
- AMD: Autoregressive Motion Diffusion (AAAI 2024 · similarity 0.89)
- Rethinking Diffusion for Text-Driven Human Motion Generation: Redundant Representations, Evaluation, and Masked Autoregression (CVPR 2025 · similarity 0.89)
- MMM: Generative Masked Motion Model (CVPR 2024 · 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.
