Efficient downsampling plays a crucial role in point cloud learning, particularly for large-scale 3D scenes. Existing downsampling methods either require a huge computational burden or sacrifice fine-grained geometric information. This paper presents an advanced sampler that achieves both high accuracy and efficiency. The proposed method utilizes voxel-based sampling as a foundation, but effectively addresses the challenges regarding voxel size determination and the preservation of critical geometric cues. Specifically, we propose a Voxel Adaptation Module that adaptively adjusts voxel sizes with the reference of point-based downsampling ratio. This ensures the sampling results exhibit a favorable distribution for comprehending various 3D objects or scenes. Additionally, we introduce a network compatible with arbitrary voxel sizes for sampling and feature extraction while maintaining high efficiency. Our method achieves state-of-the-art accuracy on the ShapeNetPart and ScanNet benchmarks with promising efficiency. Code will be available at https://github.com/yhc2021/AVS-Net.

翻译：高效下采样在点云学习中至关重要，尤其对于大规模三维场景。现有下采样方法要么计算负担巨大，要么牺牲细粒度几何信息。本文提出了一种兼具高精度与高效率的先进采样器。所提方法以基于体素的采样为基础，但有效解决了体素尺寸确定和关键几何线索保留方面的挑战。具体而言，我们提出一种体素适应模块，该模块参考基于点的下采样比例自适应调整体素尺寸，从而确保采样结果具有有利于理解各类三维物体或场景的分布。此外，我们引入了一种与任意体素尺寸兼容的网络，用于采样和特征提取，同时保持高效率。我们的方法在ShapeNetPart和ScanNet基准测试上以高效性能达到了最先进的精度。代码将在https://github.com/yhc2021/AVS-Net公开。