Research map / 3D vision / Face and shape
Visual-to-Executable Procedural Reconstruction of Buildings.
Zichong Zhu, Zhihao Liu, Andrei Sharf, Zhanglin Cheng
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Studies image-based reconstruction of buildings as executable procedural programs, aiming to replace manually designed grammars with automatically inferred rules.
Problem
Procedural building representations are compact and editable but require manually designed grammars that are hard to obtain in practice; the paper targets recovering them from images.
Approach
Image-based inference of executable procedural rules for buildings, so that reconstructed geometry is encoded as runnable procedural programs rather than static meshes. Details beyond the abstract are not stated.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| predict structured intermediate representation (middle code) instead of direct low-level CGA code (ablation) | Gemini 3.1 Pro Preview · Tile-level generation | execution success: 0.68 -> 0.95 | Improved |
| use fine-tuned component predictor instead of prompted middle-code generation (ablation) | component parameterization · Tile-level generation | execution success: 0.95 -> 0.98 | Improved |
| hierarchical pipeline versus direct end-to-end LLM CGA generation | full framework · Full-program comparison | TRC: 1.83 -> 1.47 | Improved |
| add first visual-symbolic reflection iteration (ablation) | layout induction · Layout recovery | TRC: 4.32 -> 2.01 | Improved |
| add second reflection iteration (ablation) | layout induction · Layout recovery | TRC: 2.01 -> 1.47 | Improved |
| more reflection iterations beyond two (ablation) | layout induction · Layout recovery | structural accuracy | No clear effect |
| combine CLIP and DINOv3 streams versus CLIP-only (ablation) | tile encoder · Window tile attributes | Sill Height accuracy: 0.72 -> 0.92 | Improved |
| combine CLIP and DINOv3 streams versus DINOv3-only (ablation) | tile encoder · Window tile attributes | Arch Height accuracy: 0.88 -> 0.90 | Improved |
| add real training crops to synthetic training (ablation) | component predictor · 107 real held-out tiles | overall macro accuracy: 0.868 -> 0.940 | Improved |
Closest papers by meaning
- ArcPro: Architectural Programs for Structured 3D Abstraction of Sparse Points (CVPR 2025 · similarity 0.88)
- MaskPLAN: Masked Generative Layout Planning from Partial Input (CVPR 2024 · similarity 0.84)
- HouseLayout3D: A Benchmark and Training-free Baseline for 3D Layout Estimation in the Wild (NeurIPS 2025 · similarity 0.84)
- CityDreamer: Compositional Generative Model of Unbounded 3D Cities (CVPR 2024 · similarity 0.84)
- 3DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Code (NeurIPS 2026 · similarity 0.84)
- SILSA: Sliding-Window Slice Latents for Topology-Preserving High-Resolution 3D Generation (NeurIPS 2026 · similarity 0.83)
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.
