Deep Point Cloud Normal Estimation via Triplet Learning

TL;DR

A novel normal estimation method for point clouds which consists of two phases: feature encoding to learn representations of local patches, and normal estimation that takes the learned representation as input and regresses the normal vector.

Abstract

Normal estimation on 3D point clouds is a fundamental problem in 3D vision and graphics. Current methods often show limited accuracy in predicting normals at sharp features (e.g., edges and corners) and less robustness to noise. In this paper, we propose a novel normal estimation method for point clouds. It consists of two phases: (a) feature encoding which learns representations of local patches, and (b) normal estimation that takes the learned representation as input and regresses the normal vector. We are motivated that local patches on isotropic and anisotropic surfaces have similar or distinct normals, and that separable features or representations can be learned to facilitate normal estimation. To realise this, we first construct triplets of local patches on 3D point cloud data, and design a triplet network with a triplet loss for feature encoding. We then design a simple network with several MLPs and a loss function to regress the normal vector. Despite having a smaller network size compared to most other methods, experimental results show that our method preserves sharp features and achieves better normal estimation results on CAD-like shapes.

Figures (6)

• Figure 1: The overall architecture of our method, where the patch size N is empirically set to 500.
• Figure 2: Illustration of triplet-based learning. The anchor, positive and negative patches are marked in dark blue, green and red respectively.
• Figure 3: Demonstration of synthetic shapes where: (a) for training; (b) for validation and (c) for testing.
• Figure 4: MSAE comparisons on noisy CAD shapes, where (a) and (b) demonstrate MSAE on synthetic shapes, and (c) demonstrates MSAE on scanned shapes with known ground-truth normals. The two best results are in bold in each row.
• Figure 5: MSAE comparisons with DFP on Tetrahedron and Cube with 1% noise.