Research map / Point clouds / Point cloud classification
Spherical Kernel for Efficient Graph Convolution on 3D Point Clouds
Huan Lei, Naveed Akhtar, Ajmal Saeed Mian
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 spherical SPH3D kernel (8x2x2+1 bins) instead of uniform 3x3x3 CNN3D kernel (ablation) | graph network · S3DIS Area 5 (2048 points) | mIoU: 57.9 -> 61.0 | Improved |
| use spherical kernel instead of 5x5x5 CNN3D kernel (ablation) | graph network · S3DIS Area 5 (2048 points) | mIoU: 58.6 -> 61.0 | Improved |
| add self-convolution weight w0 to 3x3x3 CNN3D kernel (ablation) | CNN3D kernel · S3DIS Area 5 (2048 points) | mIoU: 57.9 -> 58.3 | No clear effect |
| add self-convolution weight w0 to 5x5x5 CNN3D kernel (ablation) | CNN3D kernel · S3DIS Area 5 (2048 points) | mIoU: 58.6 -> 58.3 | No clear effect |
| use even-size 4x4x4 kernel (asymmetric) instead of odd 5x5x5 (ablation) | CNN3D kernel · S3DIS Area 5 (2048 points) | mIoU: 58.6 -> 60.8 | Improved |
| use max pooling instead of average pooling (ablation) | SPH3D-GCN · S3DIS Area 5 (8192 points) | mIoU: 58.8 -> 59.5 | Improved |
| add data augmentation (ablation) | SPH3D-GCN · S3DIS Area 5 (8192 points) | mIoU: 57.0 -> 59.5 | Improved |
| use weighted instead of uniform interpolation (dense, 8192 points) (ablation) | SPH3D-GCN · S3DIS Area 5 (8192 points) | mIoU: 59.5 -> 61.2 | Improved |
| use weighted instead of uniform interpolation (sparse, 4096 points) (ablation) | SPH3D-GCN · S3DIS Area 5 (4096 points) | mIoU: 62.2 -> 61.1 | Hurt |
| remove relative location from input (9-dim to 6-dim) (ablation) | SPH3D-GCN · S3DIS Area 5 | mIoU: 58.6 -> 59.5 | Improved |
| use separable instead of regular (V1) convolution (ablation) | SPH3D-GCN · ModelNet40 | instance accuracy: 91.3 -> 91.4 | No clear effect |
| use separable instead of regular (V1) convolution (ablation) | SPH3D-GCN · Random 2048 points | per-batch time (ms): 36.1 -> 14.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PointCNN — PointCNN: Convolution On X-Transformed Points (NeurIPS 2018) · better on 5 of 5 shared results
- PointConv — PointConv: Deep Convolutional Networks on 3D Point Clouds (CVPR 2019) · better on 1 of 1 shared result
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
- Octree Guided CNN With Spherical Kernels for 3D Point Clouds (CVPR 2019 · similarity 0.93)
- Convolution in the Cloud: Learning Deformable Kernels in 3D Graph Convolution Networks for Point Cloud Analysis (CVPR 2020 · similarity 0.92)
- Mining Point Cloud Local Structures by Kernel Correlation and Graph Pooling (CVPR 2018 · similarity 0.91)
- Learning of 3D Graph Convolution Networks for Point Cloud Analysis (TPAMI 2021 · similarity 0.91)
- SegGCN: Efficient 3D Point Cloud Segmentation With Fuzzy Spherical Kernel (CVPR 2020 · similarity 0.90)
- Learning Localized Generative Models for 3D Point Clouds via Graph Convolution (ICLR 2019 · similarity 0.89)
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.
