Research map / Point clouds / Point cloud classification
KPConv: Flexible and Deformable Convolution for Point Clouds
Hugues Thomas, Charles R. Qi, Jean-Emmanuel Deschaud, Beatriz Marcotegui, Francois Goulette, Leonidas J. Guibas
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 deformable kernels instead of rigid | KPConv · ModelNet40 | overall accuracy: 92.9 -> 92.7 | Hurt |
| use deformable kernels instead of rigid | KPConv · ShapeNetPart | instance mIoU: 86.2 -> 86.4 | Improved |
| use deformable kernels instead of rigid | KPConv · S3DIS Area-5 | mIoU: 65.4 -> 67.1 | Improved |
| use deformable kernels instead of rigid | KPConv · Paris-Lille-3D | mIoU: 72.3 -> 75.9 | Improved |
| use deformable kernels instead of rigid | KPConv · Scannet (test) | mIoU: 68.6 -> 68.4 | Hurt |
| use deformable kernels instead of rigid | KPConv · Semantic3D reduced-8 | mIoU: 74.6 -> 73.1 | Hurt |
| use deformable kernels instead of rigid | KPConv · S3DIS (k-fold) | mIoU: 69.6 -> 70.6 | Improved |
| reduce kernel points to 4 for deformable vs rigid (ablation) | KPConv · Scannet validation | mIoU: deformable loses 1.5% vs rigid loses 3.5% | Improved |
| use deformable instead of rigid with 15 kernel points (ablation) | KPConv · Scannet validation | mIoU | Improved |
| replace nearest upsampling by KPConv in decoder (ablation) | KP-FCNN · segmentation | mIoU | No clear effect |
| add fitting and repulsive regularization loss to deformable kernels (ablation) | deformable KPConv · segmentation | qualitative | Improved |
| use KPConv instead of other point convolutions on Scannet | KP-FCNN · Scannet | mIoU: exceed previous scores by 19 mIoU points | 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 4 of 4 shared results
- SGPN — SGPN: Similarity Group Proposal Network for 3D Point Cloud Instance Segmentation (CVPR 2018) · 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
- KPConvX: Modernizing Kernel Point Convolution with Kernel Attention (CVPR 2024 · similarity 0.91)
- FPConv: Learning Local Flattening for Point Convolution (CVPR 2020 · similarity 0.89)
- PointConv: Deep Convolutional Networks on 3D Point Clouds (CVPR 2019 · similarity 0.89)
- PAConv: Position Adaptive Convolution With Dynamic Kernel Assembling on Point Clouds (CVPR 2021 · similarity 0.88)
- PointCNN: Convolution On X-Transformed Points (NeurIPS 2018 · similarity 0.88)
- RepKPU: Point Cloud Upsampling with Kernel Point Representation and Deformation (CVPR 2024 · similarity 0.86)
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.
