Research map / Point clouds / Point cloud classification
Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework
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 |
|---|---|---|---|
| Replace specialized local feature extractors with hierarchical residual MLPs | Point cloud classification · ModelNet40 | OA: 94.2 -> 94.5 | Helped |
| Use the geometric affine module (ablation) | PointMLP · ScanObjectNN PB_T50_RS test | OA: 81.5±1.4 -> 85.4±0.3 | Helped |
| Add the pre-MLP blocks before aggregation (ablation) | PointMLP · ScanObjectNN PB_T50_RS test | OA: 82.8±0.0 -> 85.4±0.3 | Helped |
| Add the post-MLP blocks after aggregation (ablation) | PointMLP · ScanObjectNN PB_T50_RS test | OA: 84.7±0.2 -> 85.4±0.3 | Helped |
| Use 40 layers rather than 24 layers (ablation) | PointMLP · ScanObjectNN PB_T50_RS test | OA: 84.8±0.5 -> 85.4±0.3 | Helped |
| Increase depth from 40 to 56 layers (ablation) | PointMLP · ScanObjectNN PB_T50_RS test | OA: 85.4±0.3 -> 85.0±0.1 | Hurt |
| Add residual connections (ablation) | PointMLP · ModelNet40 | OA: 88.1% (6% drop) | Helped |
| Use post-MLP blocks instead of replacing them with more pre-MLP blocks (ablation) | PointMLP · ScanObjectNN PB_T50_RS test | OA: 84.13% -> 85.4% | Helped |
| Add the geometric affine module (ablation) | PointNet++ · ModelNet40 | OA: +1.4% | Helped |
| Add the geometric affine module (ablation) | DGCNN · ModelNet40 | OA: 92.9% -> 92.8% | Hurt |
| Use the PointMLP-elite architecture instead of full PointMLP | Point cloud classification · ModelNet40 | test speed: 112 -> 176 | Helped |
Closest papers by meaning
- Point Deformable Network with Enhanced Normal Embedding for Point Cloud Analysis (AAAI 2024 · similarity 0.89)
- Improved MLP Point Cloud Processing with High-Dimensional Positional Encoding (AAAI 2024 · similarity 0.89)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.88)
- A Linear Time and Space Local Point Cloud Geometry Encoder via Vectorized Kernel Mixture (VecKM) (ICML 2024 · similarity 0.88)
- PointCFormer: A Relation-Based Progressive Feature Extraction Network for Point Cloud Completion (AAAI 2025 · similarity 0.88)
- DeepLA-Net: Very Deep Local Aggregation Networks for Point Cloud Analysis (CVPR 2025 · 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.