Research map / Image generation / Text-to-image diffusion
Pose-RFT: Aligning MLLMs for 3D Pose Generation via Hybrid Action Reinforcement Fine-Tuning
Bao Li, Xiaomei Zhang, Miao Xu, Zhaoxin Fan, Xiangyu Zhu, Zhen Lei
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 Gaussian distributional pose head without RFT (ablation) | pose-specific MLLM baseline · 3DPW | MPJPE: 90.4 -> 91.4 | Hurt |
| add HyGRPO reinforcement fine-tuning on top of the Gaussian head (ablation) | Pose-RFT · 3DPW | MPJPE: 91.4 -> 85.9 | Improved |
| add HyGRPO reinforcement fine-tuning on top of the Gaussian head (ablation) | Pose-RFT · PoseScript-H2 | mRecall text-to-pose: 37.4 -> 53.6 | Improved |
| add HyGRPO reinforcement fine-tuning on top of the Gaussian head (ablation) | Pose-RFT · PoseScript-H2 | mRecall pose-to-text: 42.0 -> 57.6 | Improved |
| use HyGRPO instead of GRPO (discrete-only) (ablation) | pose-specific MLLM RFT · mean group reward during training | mean group reward | Improved |
| add pose-aware ViT encoder with separate projections (ablation) | Pose-RFT · 3DPW | joint location reward | Improved |
| add pose-aware ViT encoder (ablation) | Pose-RFT · PoseScript-H2 | semantic alignment reward | No clear effect |
| add reinforcement fine-tuning with HyGRPO (1,000 steps) (ablation) | pose-specific MLLM · PoseScript and 3DPW rewards | reward | Improved |
Closest papers by meaning
- PoseLLaVA: Pose Centric Multimodal LLM for Fine-Grained 3D Pose Manipulation (AAAI 2025 · similarity 0.88)
- Stable-Pose: Leveraging Transformers for Pose-Guided Text-to-Image Generation (NeurIPS 2024 · similarity 0.87)
- TrioPose: Native Triple-Stream Diffusion Transformers for Pose-Guided Text-to-Image Generation (NeurIPS 2026 · similarity 0.86)
- UniPose: A Unified Multimodal Framework for Human Pose Comprehension, Generation and Editing (CVPR 2025 · similarity 0.86)
- MotionMap: Representing Multimodality in Human Pose Forecasting (CVPR 2025 · similarity 0.85)
- HMVLM:Human Motion-Vision-Language Model via MoE LoRA (NeurIPS 2025 · similarity 0.85)
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.
