Research map / 3D vision / 3D generation
PlaneRAS: Learning Planar Primitives for 3D Plane Recovery
Fang Zhang, Wenzhao Zheng, Linqing Zhao, Zelan Zhu, Jiwen Lu, Xiuzhuang Zhou
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 pretrained monocular depth prior (ablation) | PlaneRAS (ResNet50) · ScanNet val | VI: 0.701 -> 0.610 | Improved |
| add pretrained depth prior (ablation) | PlaneRAS (ResNet50) · ScanNet val | plane offset error (mm): 161.56 -> 106.61 | Improved |
| add normal prior on top of depth prior (ablation) | PlaneRAS (ResNet50) · ScanNet val | VI: 0.610 -> 0.608 | No clear effect |
| add normal prior on top of depth prior (ablation) | PlaneRAS (ResNet50) · ScanNet val | normal error (deg): 7.77 -> 7.88 | No clear effect |
| use plane kernel instead of 2D Gaussian kernel (no prior) (ablation) | PlaneRAS (ResNet50) · ScanNet val | VI: 0.762 -> 0.701 | Improved |
| use plane kernel instead of 2D Gaussian kernel (no prior) (ablation) | PlaneRAS (ResNet50) · ScanNet val | SC: 0.840 -> 0.856 | Improved |
| reduce primitive resolution from 48x64 to 36x48 (ablation) | PlaneRAS (ResNet50) · ScanNet val | VI: 0.610 -> 0.630 | Hurt |
| reduce primitive resolution from 48x64 to 24x32 (ablation) | PlaneRAS (ResNet50) · ScanNet val | VI: 0.610 -> 0.653 | Hurt |
| reduce primitive resolution from 48x64 to 36x48 (ablation) | PlaneRAS (ResNet50) · ScanNet val | normal error (deg): 7.77 -> 7.40 | Improved |
| use explicit 3D planar primitives instead of PlaneRecTR 2D queries | PlaneRecTR (ResNet50) · ScanNet val | VI: 0.698 -> 0.610 | Improved |
| use explicit 3D planar primitives instead of PlaneRecTR 2D queries | PlaneRecTR (ResNet50) · ScanNet val | plane parameter errors: 10.23 -> 7.77 | Improved |
| use PlaneRAS with pretrained depth priors instead of PlaneRecTR (Swin-B) for depth | PlaneRecTR (Swin-B) · NYUv2 planar-region depth | Rel: 0.157 -> 0.165 | Hurt |
Closest papers by meaning
- PLANA3R: Zero-shot Metric Planar 3D Reconstruction via Feed-forward Planar Splatting (NeurIPS 2025 · similarity 0.89)
- PlanarSplatting: Accurate Planar Surface Reconstruction in 3 Minutes (CVPR 2025 · similarity 0.89)
- AlphaTablets: A Generic Plane Representation for 3D Planar Reconstruction from Monocular Videos (NeurIPS 2024 · similarity 0.88)
- Towards In-the-wild 3D Plane Reconstruction from a Single Image (CVPR 2025 · similarity 0.88)
- AirPlanes: Accurate Plane Estimation via 3D-Consistent Embeddings (CVPR 2024 · similarity 0.88)
- PlanarGS: High-Fidelity Indoor 3D Gaussian Splatting Guided by Vision-Language Planar Priors (NeurIPS 2025 · similarity 0.87)
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.
