Research map / Point clouds / Point cloud classification
Escape From Cells: Deep Kd-Networks for the Recognition of 3D Point Cloud Models
Roman Klokov, Victor Lempitsky
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 |
|---|---|---|---|
| randomize kd-tree generation (RT instead of deterministic DT) (ablation) | Kd-net depth 10 · ModelNet40 | accuracy: 85.7 -> 89.9 | Improved |
| randomize kd-tree generation (RT instead of DT) (ablation) | Kd-net depth 10 · ModelNet10 | accuracy: 89.2 -> 92.8 | Improved |
| add anisotropic scaling augmentation to randomized trees (ablation) | Kd-net RT depth 10 · ModelNet40 | accuracy: 89.9 -> 90.6 | Improved |
| add translation augmentation to randomized trees (ablation) | Kd-net RT depth 10 · ModelNet40 | accuracy: 89.9 -> 90.1 | Improved |
| add translation augmentation on top of RT+SA (ablation) | Kd-net RT+SA depth 10 · ModelNet40 | accuracy: 90.6 -> 90.6 | No clear effect |
| add translation augmentation on top of RT+SA (ablation) | Kd-net RT+SA depth 10 · ModelNet10 | accuracy: 93.2 -> 93.3 | No clear effect |
| add informative leaf representations (vs constant scalar leaves) (ablation) | Kd-net RT+SA depth 10 · ModelNet40 | accuracy: 89.8 -> 90.6 | Improved |
| add informative leaf representations (vs constant scalar leaves) (ablation) | Kd-net RT+SA depth 10 · ModelNet10 | accuracy: 92.7 -> 93.2 | Improved |
| replace linear classifier on one-hot split directions with hierarchical Kd-network (no leaf info) (ablation) | split-direction features · ModelNet40 | accuracy: 73.2 -> 89.8 | Improved |
| increase kd-tree depth from 10 to 15 | Kd-Net · ModelNet40 | instance accuracy: 90.6 -> 91.8 | Improved |
| increase kd-tree depth from 10 to 15 | Kd-Net · ModelNet10 | instance accuracy: 93.3 -> 94.0 | Improved |
| apply Kd-net to 20 random rotations with max-pooling (multi-view MVKd-net) | Kd-net on perturbed ShapeNetCore · ShapeNetCore perturbed (micro) | mAP: 0.617 -> 0.766 | Improved |
Closest papers by meaning
- PointGrid: A Deep Network for 3D Shape Understanding (CVPR 2018 · similarity 0.88)
- Multiresolution Tree Networks for 3D Point Cloud Processing (ECCV 2018 · similarity 0.86)
- Convolution in the Cloud: Learning Deformable Kernels in 3D Graph Convolution Networks for Point Cloud Analysis (CVPR 2020 · similarity 0.86)
- Unsupervised Deep Shape Descriptor With Point Distribution Learning (CVPR 2020 · similarity 0.86)
- Attentional ShapeContextNet for Point Cloud Recognition (CVPR 2018 · similarity 0.86)
- DeepLA-Net: Very Deep Local Aggregation Networks for Point Cloud Analysis (CVPR 2025 · similarity 0.85)
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.
