Research map / Vision-language and multimodal / Multimodal LLMs
3D-Layout-R1: Structured Reasoning for Language-Instructed Spatial Editing
Haoyu Zhen, Xiaolong Li, Yilin Zhao, Han Zhang, Sifei Liu, Kaichun Mo, Chuang Gan, Subhashree Radhakrishnan
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 structured CoT-SFT over answer-only SFT | Qwen3-8B · 3D Sorting | Mean IoU: 0.552 -> 0.635 | Improved |
| add GRPO after structured CoT-SFT | Qwen3-8B · 3D Sorting | Mean IoU: 0.635 -> 0.924 | Improved |
| add structured CoT-SFT before GRPO (vs vanilla GRPO) (ablation) | Qwen3-8B · 3D Sorting | Mean IoU: 0.850 -> 0.924 | Improved |
| add structured CoT-SFT before GRPO (vs vanilla GRPO) (ablation) | Cosmos-7B · 3D Sorting | Mean IoU: 0.729 -> 0.875 | Improved |
| add structured CoT-SFT before GRPO (vs vanilla GRPO) (ablation) | Qwen3-8B · RoomEditing | Mean IoU: 0.768 -> 0.983 | Improved |
| add GRPO after CoT-SFT | Qwen2.5-7B · RoomEditing | Mean IoU: 0.950 -> 0.974 | Improved |
| add GRPO after CoT-SFT | Qwen3-8B · RoomEditing | Mean IoU: 0.972 -> 0.983 | No clear effect |
| add GRPO after CoT-SFT | Qwen2.5-VL-7B · Sorting caption-free | IoU: 0.641 -> 0.787 | Improved |
| add GRPO after CoT-SFT | Qwen2.5-VL-7B · Sorting noisy input | IoU: 0.773 -> 0.822 | Improved |
| add format reward (ablation) | Qwen3-8B GRPO · 3D Sorting | IoU: 0.819 -> 0.924 | Improved |
| add IoU reward (ablation) | Qwen3-8B GRPO · 3D Sorting | IoU: 0.415 -> 0.924 | Improved |
| add collision-free reward (ablation) | Qwen3-8B GRPO · 3D Sorting | IoU: 0.919 -> 0.924 | Improved |
Closest papers by meaning
- LayoutVLM: Differentiable Optimization of 3D Layout via Vision-Language Models (CVPR 2025 · similarity 0.90)
- 3D Space as a Scratchpad for Editable Text-to-Image Generation (CVPR 2026 · similarity 0.90)
- Think, Plan, Paint: Layout-Aware Reasoning for Controllable Image Generation in Unified Models (NeurIPS 2026 · similarity 0.89)
- SKE-Layout: Spatial Knowledge Enhanced Layout Generation with LLMs (CVPR 2025 · similarity 0.89)
- Scene Graph Thinking: Reinforcing Structured Visual Reasoning for Multimodal Large Language Models (ICML 2026 · similarity 0.88)
- InstructScene: Instruction-Driven 3D Indoor Scene Synthesis with Semantic Graph Prior (ICLR 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.
