Abstract:Point cloud representation has gained traction due to its efficient memory usage and simplicity in acquisition, manipulation, and storage. However, as point cloud sizes increase, effective down-sampling becomes essential to address the computational requirements of downstream tasks. Classical approaches, such as furthest point sampling (FPS), perform well on benchmarks but rely on heuristics and overlook geometric features, like curvature, during down-sampling. In this paper, We introduce a reinforcement learning-based sampling algorithm that enhances FPS by integrating curvature information. Our approach ranks points by combining FPS-derived soft ranks with curvature scores computed by a deep neural network, allowing us to replace a proportion of low-curvature points in the FPS set with high-curvature points from the unselected set. Existing differentiable sampling techniques often suffer from training instability, hindering their integration into end-to-end learning frameworks. By contrast, our method achieves stable end-to-end learning, consistently outperforming baseline models across multiple downstream geometry processing tasks. We provide comprehensive ablation studies, with both qualitative and quantitative insights into the effect of each feature on performance. Our algorithm establishes state-of-the-art results for classification, segmentation and shape completion, showcasing its robustness and adaptability.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of effective down - sampling in large - scale applications of point cloud data. As the amount of point cloud data increases, the demand for computing resources also increases. Therefore, effective down - sampling methods are required to reduce the amount of point cloud data while retaining its geometric features and structural information to meet the computing requirements of downstream tasks (such as classification, segmentation and shape completion). #### Specific problems: 1. **Limitations of existing down - sampling methods**: - **Furthest Point Sampling (FPS)**: Although FPS performs well in benchmark tests, it relies on heuristic methods and ignores geometric features (such as curvature) in the point cloud. This may lead to the loss of important details in the down - sampled point cloud. - **Non - differentiability and instability**: Existing differentiable down - sampling techniques often have problems of unstable training and are difficult to integrate into end - to - end learning frameworks. 2. **Task - specific down - sampling requirements**: - Existing methods are usually task - independent and cannot prioritize points that are crucial for specific downstream tasks. For example, in 3D reconstruction tasks, points with high curvature but low density are very important, but existing methods cannot effectively identify these points. #### Solutions proposed in the paper: To solve the above problems, the authors propose a down - sampling algorithm based on reinforcement learning - **Curvature - Informed Furthest Point Sampling (CFPS)**. This method dynamically adjusts the selection of down - sampling points by combining the soft ranking generated by FPS and the curvature scores calculated by the deep neural network. Specifically: - **Introducing curvature information**: Calculate the curvature value of each point and combine it with the soft ranking generated by FPS to form a joint ranking. This can ensure that high - curvature points are preferentially retained during the down - sampling process, thereby better capturing the local geometric features of the point cloud. - **Dynamically adjusting the exchange ratio**: Use the policy gradient estimator to dynamically adjust the exchange ratio between the core point set and the non - core point set and select the optimal points for replacement. - **End - to - end learning compatibility**: This method can run stably in the end - to - end learning framework and is superior to traditional and differentiable down - sampling methods. Through these improvements, CFPS can achieve better performance in multiple downstream geometric processing tasks, especially showing superior performance in classification, segmentation and shape completion tasks. ### Summary The core problem of the paper is to improve the effectiveness of point cloud down - sampling and the ability to retain geometric features, especially for points in high - curvature regions. By introducing curvature information and dynamically adjusting the down - sampling strategy, CFPS not only solves the limitations of existing methods but also achieves significant performance improvements in multiple tasks.

Curvature Informed Furthest Point Sampling

Curvature-aware adaptive re-sampling for point-sampled geometry

Simplification of Point-Sampled Geometry with Feature Preservation

Simplification of Point-Sampled Model Based on Geometry Images

Curvature estimation of point-sampled surfaces and its applications

Enhancing Surface Neural Implicits with Curvature-Guided Sampling and Uncertainty-Augmented Representations

Visualizing Point Cloud Classifiers by Curvature Smoothing

Curvature Regularized Surface Reconstruction from Point Cloud

An Adjustable Farthest Point Sampling Method for Approximately-sorted Point Cloud Data

Efficient Method For Geometric Attribute Estimation

Error-aware Sampling in Adaptive Shells for Neural Surface Reconstruction

Robust Estimation of Surface Curvature Information from Point Cloud Data

PLATYPUS: Progressive Local Surface Estimator for Arbitrary-Scale Point Cloud Upsampling

CAP-UDF: Learning Unsigned Distance Functions Progressively from Raw Point Clouds with Consistency-Aware Field Optimization

A Density-Aware Point Cloud Geometry Compression Leveraging Cluster- Centric Processing

PFGS: High Fidelity Point Cloud Rendering via Feature Splatting

A Closer Look at Few-Shot 3D Point Cloud Classification

FPSMix: data augmentation strategy for point cloud classification

Data-driven Upsampling of Point Clouds

Concavity-Induced Distance for Unoriented Point Cloud Decomposition

Curvy: A Parametric Cross-section based Surface Reconstruction