Research map / 3D vision / 3D generation
InstructLayout: Instruction-Driven 2D and 3D Layout Synthesis With Semantic Graph Prior
Chenguo Lin, Yu-Chen Lin, Panwang Pan, Xuanyang Zhang, Yadong Mu
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 independent-mask discrete diffusion graph prior instead of Gaussian diffusion (ablation) | InstructLayout · 3D Bedroom | iRecall (%): 34.18 -> 73.64 | Improved |
| add independent masking instead of joint masking of nodes and attributes (ablation) | InstructLayout · 3D Bedroom | iRecall (%): 34.21 -> 73.64 | Improved |
| add mask states instead of uniform transition matrices (ablation) | InstructLayout · 3D Bedroom | FID: 139.61 -> 114.78 | Improved |
| use random-shuffled non-invariant graph modeling instead of permutation-invariant (ablation) | InstructLayout · 3D Bedroom | iRecall (%): 70.49 -> 73.64 | Improved |
| use permutation non-invariant graph modeling (ablation) | InstructLayout · 3D Bedroom | FID: 116.53 -> 114.78 | Improved |
| reduce graph-prior timesteps from 100 to 50 (ablation) | InstructLayout · 2D poster | iRecall (%): 78.268 -> 78.833 | No clear effect |
| reduce graph-prior timesteps from 100 to 25 (ablation) | InstructLayout · 2D poster | iRecall (%): 78.268 -> 77.862 | No clear effect |
| reduce LayoutDM timesteps from 110 to 35 (ablation) | LayoutDM · 2D poster | iRecall (%): 65.841 -> 61.825 | Hurt |
| use two-stage graph prior + layout decoder instead of ATISS | ATISS · 3D Bedroom | iRecall (%): 48.13 -> 73.64 | Improved |
| use two-stage graph prior + layout decoder instead of DiffuScene | DiffuScene · 3D Living room | FID: 122.20 -> 110.39 | Improved |
| use two-stage discrete+continuous diffusion instead of single-stage LayoutDM | LayoutDM · 2D poster | iRecall (%): 62.999 -> 78.451 | Improved |
| use two-stage diffusion instead of continuous DDPM | DDPM · 2D poster | Ove x100: 17.573 -> 0.785 | 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.
- DiffuScene — DiffuScene: Denoising Diffusion Models for Generative Indoor Scene Synthesis (CVPR 2024) · better on 6 of 6 shared results
- ATISS — ATISS: Autoregressive Transformers for Indoor Scene Synthesis (NeurIPS 2021) · better on 6 of 6 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
- InstructScene: Instruction-Driven 3D Indoor Scene Synthesis with Semantic Graph Prior (ICLR 2024 · similarity 0.95)
- Direct Numerical Layout Generation for 3D Indoor Scene Synthesis via Spatial Reasoning (NeurIPS 2025 · similarity 0.91)
- Layout-your-3D: Controllable and Precise 3D Generation with 2D Blueprint (ICLR 2025 · similarity 0.90)
- LayoutVLM: Differentiable Optimization of 3D Layout via Vision-Language Models (CVPR 2025 · similarity 0.89)
- Rethinking The Training And Evaluation of Rich-Context Layout-to-Image Generation (NeurIPS 2024 · similarity 0.89)
- Think, Plan, Paint: Layout-Aware Reasoning for Controllable Image Generation in Unified Models (NeurIPS 2026 · 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.
