Research map / Point clouds / Shape and surface completion
Shape As Points: A Differentiable Poisson Solver
Songyou Peng, Chiyu Max Jiang, Yiyi Liao, Michael Niemeyer, Marc Pollefeys, Andreas Geiger
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 differentiable Poisson solver (SAP) vs SPSR | optimization-based reconstruction · Thingi10K | Chamfer-L1: 0.223 -> 0.054 | Improved |
| add DPSR vs IGR | optimization-based reconstruction · SRB | Chamfer-L1: 0.178 -> 0.076 | Improved |
| add DPSR vs SPSR | optimization-based reconstruction · D-FAUST | Chamfer-L1: 0.044 -> 0.043 | No clear effect |
| remove DPSR loss on indicator grid, supervise points with Chamfer only (ablation) | SAP · ShapeNet noise=0.005, outliers=50% | Chamfer-L1: 0.038 -> 0.085 | Hurt |
| remove DPSR loss (ablation) | SAP · ShapeNet noise=0.005 | Chamfer-L1: 0.034 -> 0.044 | Hurt |
| remove DPSR loss (ablation) | SAP · ShapeNet noise=0.025 | Chamfer-L1: 0.054 -> 0.067 | Hurt |
| increase offsets per input point from 1x to 7x (ablation) | SAP · ShapeNet outliers setup | Chamfer-L1: 0.041 -> 0.038 | Improved |
| replace 2D encoder with 3D encoder (ablation) | SAP · ShapeNet outliers setup | Chamfer-L1: 0.043 -> 0.038 | Improved |
| replace ConvONet with SAP | ShapeNet reconstruction · ShapeNet noise=0.005 | runtime (s) at 128^3: 0.327 -> 0.064 | Improved |
| remove point resampling (ablation) | SAP · D-FAUST | Chamfer-L1: 0.044 -> 0.043 | No clear effect |
| use geometric initialization vs ConvONet | SAP · ShapeNet training 10K iterations | Chamfer-L1: 0.082 -> 0.041 | Improved |
Closest papers by meaning
- $p$-Poisson surface reconstruction in curl-free flow from point clouds (NeurIPS 2023 · similarity 0.91)
- Shape Reconstruction by Learning Differentiable Surface Representations (CVPR 2020 · similarity 0.90)
- GridPull: Towards Scalability in Learning Implicit Representations from 3D Point Clouds (ICCV 2023 · similarity 0.89)
- POCO: Point Convolution for Surface Reconstruction (CVPR 2022 · similarity 0.88)
- DiGS: Divergence Guided Shape Implicit Neural Representation for Unoriented Point Clouds (CVPR 2022 · similarity 0.88)
- High-Fidelity Lightweight Mesh Reconstruction from Point Clouds (CVPR 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.
