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

Abstract

Sampling is an essential part of raw point cloud data processing, such as in the popular PointNet++ scheme. Farthest Point Sampling (FPS), which iteratively samples the farthest point and performs distance updating, is one of the most popular sampling schemes. Unfortunately, it suffers from low efficiency and can become the bottleneck of point cloud applications. We propose adjustable FPS (AFPS) to aggressively reduce the complexity of FPS without compromising the sampling performance. AFPS, parameterized by M, divides the original point cloud into <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$M$</tex> small point clouds and samples <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$M$</tex> points simultaneously. It exploits the dimensional locality of an approximately sorted point cloud data to minimize its performance degradation. On a multi-core platform, AFPS method with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$M$</tex> = 32 can achieve 30× speedup over original FPS. Furthermore, we propose the nearest-point-distance-updating (NPDU) method to limit the number of distance updates to a constant number. On the ShapeNet part segmentation task, using the NPDU method on AFPS on a point cloud with 2K-32K points helps achieve a 34-280× speedup with 0.8490 instance average mIoU (mean Intersection of Union), which is only 0.0035 lower than that of the original FPS.