Research map / Video / Video generation
IRG-MotionLLM: Interleaving Motion Generation, Assessment and Refinement for Text-to-Motion Generation
Yuanming Li, Qize Yang, Nan Lei, Shenghao Fu, Ling-An Zeng, Jian-Fang Hu, Xihan Wei, WEISHI ZHENG
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 assessment and refinement tasks to basic tasks in Stage-1 (ablation) | IRG-MotionLLM Stage-1 · HumanML3D T2M | R-Precision Top-1: 0.500 -> 0.504 | Improved |
| add assessment and refinement tasks to basic tasks in Stage-1 (ablation) | IRG-MotionLLM Stage-1 · HumanML3D T2M | FID: 0.123 -> 0.141 | Hurt |
| fine-tune base model on T2M task only in Stage-1 (ablation) | MotionLLM* · HumanML3D T2M | FID: 0.198 -> 0.185 | Improved |
| fine-tune base model on T2M task only in Stage-1 (ablation) | MotionLLM* · HumanML3D T2M | R-Precision Top-3: 0.782 -> 0.778 | Hurt |
| add assessment and refinement tasks in Stage-1 (ablation) | IRG-MotionLLM Stage-1 · HumanML3D motion caption | CIDEr: 48.94 -> 52.4 | Improved |
| add assessment and refinement to Stage-2 CoT data (ablation) | IRG-MotionLLM Stage-2 · HumanML3D T2M | R-Precision Top-1: 0.503 -> 0.526 | Improved |
| add Ignore Incorrect loss masking in Stage-2 (ablation) | IRG-MotionLLM Stage-2 · HumanML3D T2M | FID: 0.592 -> 0.111 | Improved |
| add Ignore Incorrect loss masking in Stage-2 (ablation) | IRG-MotionLLM Stage-2 · HumanML3D T2M | R-Precision Top-1: 0.448 -> 0.526 | Improved |
| add GRPO-based IRMoGen reinforcing (Stage-3) | IRG-MotionLLM · HumanML3D T2M | R-Precision Top-1: 0.526 -> 0.535 | Improved |
| add GRPO-based IRMoGen reinforcing (Stage-3) | IRG-MotionLLM · HumanML3D T2M | FID: 0.111 -> 0.242 | Hurt |
| add manual filtering of LLM annotations (ablation) | IRG-MotionLLM Stage-1 · HumanML3D T2M | FID: 0.152 -> 0.135 | Improved |
| scale up Stage-3 data and extend GRPO from 900 to 2,600 steps | IRG-MotionLLM Stage-3 · HumanML3D T2M | R-Precision Top-1: 0.535 -> 0.564 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MotionLLM* (base) — MotionLLM: Understanding Human Behaviors from Human Motions and Videos (TPAMI 2025) · better on 6 of 6 shared results
- Motion-R1 — Motion-R1: Enhancing Motion Generation with Decomposed Chain-of-Thought and RL Binding (ICLR 2026) · better on 2 of 2 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
- Motion-R1: Enhancing Motion Generation with Decomposed Chain-of-Thought and RL Binding (ICLR 2026 · similarity 0.90)
- MotionMaster: Generalizable Text-Driven Motion Generation and Editing (CVPR 2026 · similarity 0.90)
- AvatarGPT: All-in-One Framework for Motion Understanding Planning Generation and Beyond (CVPR 2024 · similarity 0.89)
- MG-MotionLLM: A Unified Framework for Motion Comprehension and Generation across Multiple Granularities (CVPR 2025 · similarity 0.89)
- Motion-Agent: A Conversational Framework for Human Motion Generation with LLMs (ICLR 2025 · similarity 0.88)
- ReMoGPT: Part-Level Retrieval-Augmented Motion-Language Models (AAAI 2025 · 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.
