Research map / Image generation / Text-to-image diffusion
MotionRFT: Unified Reinforcement Fine-Tuning for Text-to-Motion Generation
Xiaofeng Tan, Wanjiang Weng, Hongsong Wang, Fang Zhao, Xin Geng, Liang Wang
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 |
|---|---|---|---|
| fine-tune MLD with EasyTune (step-wise reward) | MLD · HumanML3D | FID: 0.450 -> 0.132 | Improved |
| add full MotionRFT (MotionReward rewards) on top of EasyTune | MLD · HumanML3D | FID: 0.132 -> 0.101 | Improved |
| apply MotionRFT to joint-based ACMDM | ACMDM · HumanML3D | FID: 0.109 -> 0.084 | Improved |
| apply MotionRFT to rotation-based HY-Motion | HY-Motion · HumanML3D | FID: 1.258 -> 0.985 | Improved |
| apply MotionRFT to rotation-based HY-Motion | HY-Motion · HumanML3D | R-Precision Top 1: 0.445 -> 0.499 | Improved |
| fine-tune with EasyTune/MotionRFT (semantic alignment sharpens per-prompt consistency) | MLD · HumanML3D | MultiModality: 2.267 -> 1.621 | Hurt |
| use unified MotionReward instead of representation-specific TMR/ReAlign retrieval | ReAlign · HumanML3D retrieval, rotation | Avg. recall: 82.01 -> 85.17 | Improved |
| use unified MotionReward instead of per-representation MotionCritic | MotionCritic · MotionCritic test, kinematic | accuracy: 85.08 -> 86.09 | Improved |
| step-wise EasyTune instead of trajectory-level ReFL-10 | MLD · HumanML3D | FID: 0.207 -> 0.132 | Improved |
| EasyTune instead of ReFL-10 on memory | MLD · HumanML3D | peak memory (GB): 22.10 -> 22.10 | No clear effect |
| EasyTune one-step ODE prediction reward on MLD++ | MLD++ · HumanML3D | FID: 0.073 -> 0.073 | No clear effect |
| EasyTune on MDM (SDE, noise-aware reward) | MDM · HumanML3D | FID: 0.489 -> 0.411 | 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.
- ReAlign — ReAlign: Generalizable Image Forgery Detection via Reasoning-Aligned Representation (CVPR 2026) · better on 1 of 1 shared result
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
- AToM: Aligning Text-to-Motion Model at Event-Level with GPT-4Vision Reward (CVPR 2025 · similarity 0.90)
- DartControl: A Diffusion-Based Autoregressive Motion Model for Real-Time Text-Driven Motion Control (ICLR 2025 · similarity 0.89)
- MG-MotionLLM: A Unified Framework for Motion Comprehension and Generation across Multiple Granularities (CVPR 2025 · similarity 0.89)
- Direct Post-Training Preference Alignment for Multi-Agent Motion Generation Model Using Implicit Feedback from Pre-training Demonstrations (ICLR 2025 · similarity 0.89)
- UniMo: Unified Motion Generation and Understanding with Chain of Thought (AAAI 2026 · similarity 0.88)
- OMG: Towards Open-vocabulary Motion Generation via Mixture of Controllers (CVPR 2024 · similarity 0.88)
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.
