Research map / Large language models / Instruction and fine-tuning
3D-RFT: Reinforcement Fine-Tuning for Video-based 3D Scene Understanding
Xiongkun Linghu, Jiangyong Huang, Baoxiong Jia, Siyuan Huang
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 RL stage after SFT (with VGGT prior) (ablation) | VG LLM-4B · ScanRefer | Acc@0.25: 36.4 -> 42.9 | Improved |
| add RL stage after SFT (with VGGT prior) (ablation) | VG LLM-4B · ScanRefer | Acc@0.5: 11.8 -> 15.9 | Improved |
| add RL stage after SFT (no 3D prior) (ablation) | Qwen2.5-VL-3B SFT · ScanRefer | Acc@0.25: 31.9 -> 38.2 | Improved |
| continue SFT instead of RL (no 3D prior) (ablation) | Qwen2.5-VL-3B SFT · ScanRefer | Acc@0.25: 31.9 -> 34.2 | Improved |
| add VGGT 3D prior (SFT) (ablation) | Qwen2.5-VL-3B · ScanRefer | Acc@0.25: 31.9 -> 36.4 | Improved |
| add VGGT 3D prior (SFT->RL) (ablation) | Qwen2.5-VL-3B · ScanRefer | Acc@0.25: 38.2 -> 42.9 | Improved |
| add RL training, 4-frame | VG LLM-4B · ScanNetDetection | F1@0.25: 38.2 -> 43.7 | Improved |
| add RL training, 4-frame | VG LLM-4B · ScanNetDetection | precision@0.25: 41.7 -> 54.2 | Improved |
| add RL training, 4-frame | VG LLM-4B · ScanNetDetection | recall@0.25: 35.7 -> 38.2 | Improved |
| add RL training, 6-frame | VG LLM-4B · ScanNetDetection | F1@0.25: 36.4 -> 41.7 | Improved |
| add RL training, 6-frame | VG LLM-4B · ScanNetDetection | precision@0.25: 39.7 -> 53.4 | Improved |
| add RL training, 4-frame, bin class | VG LLM-4B · ScanNetDetection | bin AP: 39.2 -> 39.1 | No clear effect |
Closest papers by meaning
- Reason-RFT: Reinforcement Fine-Tuning for Visual Reasoning of Vision Language Models (NeurIPS 2025 · similarity 0.91)
- MetaSpatial: Reinforcing 3D Spatial Reasoning in VLMs for the Metaverse (ICLR 2026 · similarity 0.89)
- VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026 · similarity 0.89)
- Spatial-SSRL: Enhancing Spatial Understanding via Self-Supervised Reinforcement Learning (CVPR 2026 · similarity 0.89)
- GPT4Scene: Understand 3D Scenes from Videos with Vision-Language Models (ICLR 2026 · similarity 0.89)
- RewardMap: Tackling Sparse Rewards in Fine-grained Visual Reasoning via Multi-Stage Reinforcement Learning (ICLR 2026 · 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.
