Research map / Point clouds / Point cloud classification
Starting From Non-Parametric Networks for 3D Point Cloud Analysis
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 sinusoidal positional encoding instead of no embedding, NeRF embedding, or Fourier embedding (ablation) | Point-NN · ModelNet40 | accuracy (%): 68.6; 70.1; 76.9 -> 81.8 | Helped |
| Combine addition and multiplication for local geometry weighting (ablation) | Point-NN · ModelNet40 | accuracy (%): 77.8; 78.3; 80.4 -> 81.8 | Helped |
| Use cosine similarity for memory-bank matching rather than Chebyshev, Euclidean, or Manhattan distance (ablation) | Point-NN · ModelNet40 | accuracy (%): 65.9; 79.8; 80.9 -> 81.8 | Helped |
| Use the memory bank instead of a decision tree, gradient boosting, or SVM classifier (ablation) | Point-NN · ModelNet40 | accuracy (%): 57.8; 63.9; 79.9 -> 81.8 | Helped |
| Insert linear layers into Point-NN to construct Point-PN (ablation) | Point-NN · ModelNet40 | accuracy (%): 81.8 -> 93.8 | Helped |
| Use Point-PN with 1+2 inserted linear layers rather than the reported 0+1, 1+1, or 2+2 variants (ablation) | Point-PN · ModelNet40 | accuracy (%): 93.4; 93.2; 92.9 -> 93.8 | Mixed |
| Do not normalize object point coordinates (ablation) | Point-NN object detection head · ScanNetV2 | AP25 (%): 4.5 -> 23.1; 7.7 -> 33.3 | Helped |
| Fuse Point-NN predictions with trained model predictions at inference | PointNet, PointNet++, PointMLP, and other pretrained models · Shape classification benchmarks | accuracy (%): +2.2; +1.2; +2.0; +0.4; +0.5; +0.2; +0.3 | Helped |
| Fuse Point-NN predictions with trained-model predictions at inference | CurveNet · ShapeNetPart | mIoU (%): +0.07 | Helped |
| Fuse Point-NN predictions with trained-model predictions at inference | VoteNet and 3DETR-m · ScanNetV2 | AR25 (%): +5.31; +11.05 | Helped |
| Use the training-free Point-NN encoder and memory bank in few-shot classification | Point-NN · ModelNet40 few-shot settings | accuracy (%): +23.4; +22.3; +33.3; +34.9 | Helped |
Closest papers by meaning
- Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework (ICLR 2022 · similarity 0.91)
- Convolution in the Cloud: Learning Deformable Kernels in 3D Graph Convolution Networks for Point Cloud Analysis (CVPR 2020 · similarity 0.89)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.88)
- ViewNet: A Novel Projection-Based Backbone With View Pooling for Few-Shot Point Cloud Classification (CVPR 2023 · similarity 0.88)
- Revisiting Point Cloud Shape Classification with a Simple and Effective Baseline (ICML 2021 · similarity 0.88)
- An Explicit Frame Construction for Normalizing 3D Point Clouds (ICML 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.