Research map / 3D vision / 3D generation
Multi-Plane Program Induction with 3D Box Priors
Yikai Li, Jiayuan Mao, Xiuming Zhang, Bill Freeman, Josh Tenenbaum, Noah Snavely, Jiajun Wu
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 box-program induction with 3D box priors instead of PlaneRCNN | BPI · Corridor plane segmentation | IoU: 0.52 -> 0.84 | Improved |
| use BPI instead of Huang et al. for plane segmentation | BPI · Corridor plane segmentation | IoU: 0.30 -> 0.84 | Improved |
| use BPI instead of Huang et al. for building plane segmentation | BPI · Building plane segmentation (original) | IoU: 0.73 -> 0.86 | Improved |
| use BPI instead of PlaneRCNN for building plane segmentation | BPI · Building plane segmentation (original) | IoU: 0.63 -> 0.86 | Improved |
| corrupt the input image before segmentation | BPI · Plane segmentation | IoU: 0.84 -> 0.84 | No clear effect |
| BPI-guided PatchMatch instead of plain PatchMatch for inpainting | PatchMatch · Corridor inpainting | L1 mean: 53.12 -> 38.45 | Improved |
| BPI-guided PatchMatch instead of GatedConv for inpainting | GatedConv · Corridor inpainting | PSNR: 32.32 -> 34.21 | Improved |
| BPI instead of GatedConv on corridor LPIPS | GatedConv · Corridor inpainting | LPIPS: 0.0153 -> 0.0170 | Hurt |
| BPI instead of Huang et al. on building inpainting | Huang et al. · Building inpainting | L1 mean: 26.10 -> 26.43 | No clear effect |
| BPI instead of GatedConv on building inpainting | GatedConv · Building inpainting | PSNR: 29.54 -> 34.86 | Improved |
| use box prior to reduce program search space (ablation) | BPI · Search time on corridor images | speedup: 23x faster | Improved |
| BPI instead of other extrapolation methods in user ranking | BPI · Building image extrapolation | preference (%): 61% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PlaneRCNN — PlaneRCNN: 3D Plane Detection and Reconstruction From a Single Image (CVPR 2019) · better on 4 of 4 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
- Perspective Plane Program Induction From a Single Image (CVPR 2020 · similarity 0.89)
- Thinking in Boxes: 3D Editing in Real Images Made Easy (NeurIPS 2026 · similarity 0.87)
- Neural Groundplans: Persistent Neural Scene Representations from a Single Image (ICLR 2023 · similarity 0.85)
- From Programs to Poses: Factored Real-World Scene Generation via Learned Program Libraries (NeurIPS 2025 · similarity 0.85)
- ArcPro: Architectural Programs for Structured 3D Abstraction of Sparse Points (CVPR 2025 · similarity 0.83)
- $\textit{Bifr\"ost}$: 3D-Aware Image Compositing with Language Instructions (NeurIPS 2024 · 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.
