Research map / Point clouds / Point cloud classification
Local Spectral Graph Convolution for Point Set Feature Learning
Chu Wang, Babak Samari, Kaleem Siddiqi
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 |
|---|---|---|---|
| replace point-MLP with spectral graph convolution on local kNN graph (ablation) | 4l-pointnet++ · ModelNet40 (1k xyz) | accuracy: 90.6 -> 91.2 | Improved |
| replace max pooling with recursive cluster pooling (ablation) | 4l-spec-max · ModelNet40 (1k xyz) | accuracy: 91.2 -> 91.5 | Improved |
| replace max pooling with recursive cluster pooling (ablation) | 4l-spec-max · ModelNet40 (2k xyz + normals) | accuracy: 91.9 -> 92.1 | Improved |
| add surface normals to xyz input (ablation) | 4l-spec-cp · ModelNet40 | accuracy: 91.5 -> 91.8 | Improved |
| increase input points from 1k to 2k (with normals) (ablation) | 4l-spec-cp · ModelNet40 | accuracy: 91.8 -> 92.1 | Improved |
| use 3 layers instead of 4 layers (ablation) | pointnet++ · ModelNet40 (1k xyz) | accuracy: 90.6 -> 90.7 | Improved |
| add spectral graph convolution and cluster pooling to pointnet++ (ablation) | 3l-pointnet++ · ShapeNet part seg (2k xyz + normals) | mIoU: 84.9 -> 85.4 | Improved |
| add spectral graph convolution and cluster pooling to pointnet++ (ablation) | 3l-pointnet++ · ShapeNet part seg (1k xyz) | mIoU: 84.2 -> 84.6 | Improved |
| replace max pooling with recursive cluster pooling (ablation) | 4l-spec · McGill Shape Benchmark (combined) | instance accuracy: 95.14 -> 95.83 | Improved |
| replace point-MLP with spectral convolution (ablation) | 4l-pointnet++ · McGill Shape Benchmark (combined) | instance accuracy: 92.36 -> 95.14 | Improved |
| replace point-MLP with spectral graph convolution plus cluster pooling (ablation) | 3l-pointnet++ · MNIST | error rate (%): 0.55 -> 0.42 | Improved |
| use spectral graph convolution plus cluster pooling (ablation) | 4l-pointnet++ · ScanNet | accuracy: 84.0 -> 84.8 | Improved |
Closest papers by meaning
- Simple Spectral Graph Convolution (ICLR 2021 · similarity 0.89)
- Mining Point Cloud Local Structures by Kernel Correlation and Graph Pooling (CVPR 2018 · similarity 0.89)
- Learning Localized Generative Models for 3D Point Clouds via Graph Convolution (ICLR 2019 · similarity 0.89)
- Long and Short-Range Dependency Graph Structure Learning Framework on Point Cloud (TPAMI 2023 · similarity 0.88)
- HoloNets: Spectral Convolutions do extend to Directed Graphs (ICLR 2024 · similarity 0.88)
- DHGCN: Dynamic Hop Graph Convolution Network for Self-Supervised Point Cloud Learning (AAAI 2024 · 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.
