Research map / Point clouds / Point cloud classification
BuildingGPT: Auto-Regressive Building Wireframe Reconstruction Model with Reinforcement Learning
Yuzhou Liu, Lingjie Zhu, Hanqiao Ye, Yujun Liu, Shangfeng Huang, Xiang Gao, Ruisheng Wang, Shuhan Shen
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 |
|---|---|---|---|
| use hierarchical footprint-to-roof tokenization instead of vanilla z-y-x ordering (ablation) | BuildingGPT pre-trained · MunichWF test | WED (x10^-2): 1.39 -> 1.19 | Improved |
| use hierarchical tokenization (ablation) | BuildingGPT pre-trained · MunichWF test | Edge F1 (%): 91.9 -> 93.1 | Improved |
| add DPO post-training with preference pairs (ablation) | BuildingGPT · MunichWF test | WED (x10^-2): 1.19 -> 0.98 | Improved |
| add DPO post-training (ablation) | BuildingGPT · MunichWF test | Edge F1 (%): 93.1 -> 94.4 | Improved |
| replace diffusion/detection with auto-regressive wireframe generation (vs EdgeDiff) | BuildingGPT · MunichWF test | WED (x10^-2): 1.39 -> 0.98 | Improved |
| scale model from 129M to 730M parameters and data from 20% to 100% (ablation) | BuildingGPT · MunichWF test cross-entropy loss | test CE loss | Improved |
| remove 25% of input points (ablation) | BuildingGPT · MunichWF test | WED (x10^-2): 0.98 -> 1.09 | No clear effect |
| remove 75% of input points (ablation) | BuildingGPT · MunichWF test | Edge F1 (%): 94.4 -> 76.6 | Hurt |
| add Gaussian noise 0.02 to input points (ablation) | BuildingGPT · MunichWF test | WED (x10^-2): 0.98 -> 1.33 | No clear effect |
| add Gaussian noise 0.05 to input points (ablation) | BuildingGPT · MunichWF test | Edge F1 (%): 94.4 -> 76.3 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- EdgeDiff — EdgeDiff: Edge-aware Diffusion Network for Building Reconstruction from Point Clouds (CVPR 2025) · better on 5 of 5 shared results
- PBWR — PBWR: Parametric-Building-Wireframe Reconstruction from Aerial LiDAR Point Clouds (CVPR 2024) · better on 4 of 4 shared results
- BWFormer — BWFormer: Building Wireframe Reconstruction from Airborne LiDAR Point Cloud with Transformer (CVPR 2025) · better on 2 of 2 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
- EdgeDiff: Edge-aware Diffusion Network for Building Reconstruction from Point Clouds (CVPR 2025 · similarity 0.85)
- cadrille: Multi-modal CAD Reconstruction with Reinforcement Learning (ICLR 2026 · similarity 0.83)
- Breaking the SFT Plateau: Multimodal Structured Reinforcement Learning for Chart-to-Code Generation (ICLR 2026 · similarity 0.83)
- ArcPro: Architectural Programs for Structured 3D Abstraction of Sparse Points (CVPR 2025 · similarity 0.83)
- From Intent to Execution: Multimodal Chain-of-Thought Reinforcement Learning for Precise CAD Code Generation (AAAI 2026 · similarity 0.82)
- GenNBV: Generalizable Next-Best-View Policy for Active 3D Reconstruction (CVPR 2024 · similarity 0.82)
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.
