PointHR: Exploring High-Resolution Architectures for 3D Point Cloud Segmentation

TL;DR

This paper introduces PointHR, a high-resolution architecture for 3D point cloud segmentation that maintains high-resolution features throughout processing, leading to improved performance over existing methods.

Contribution

It generalizes high-resolution architectures for 3D point cloud analysis using a unified pipeline with novel operators and pre-computation techniques, achieving state-of-the-art results.

Findings

PointHR outperforms recent methods on S3DIS and ScanNetV2 datasets.

Pre-computing indices reduces on-the-fly computation costs.

Maintaining high-resolution features improves segmentation accuracy.

Abstract

Significant progress has been made recently in point cloud segmentation utilizing an encoder-decoder framework, which initially encodes point clouds into low-resolution representations and subsequently decodes high-resolution predictions. Inspired by the success of high-resolution architectures in image dense prediction, which always maintains a high-resolution representation throughout the entire learning process, we consider it also highly important for 3D dense point cloud analysis. Therefore, in this paper, we explore high-resolution architectures for 3D point cloud segmentation. Specifically, we generalize high-resolution architectures using a unified pipeline named PointHR, which includes a knn-based sequence operator for feature extraction and a differential resampling operator to efficiently communicate different resolutions. Additionally, we propose to avoid numerous on-the-fly computations of high-resolution architectures by pre-computing the indices for both sequence and resampling operators. By doing so, we deliver highly competitive high-resolution architectures while capitalizing on the benefits of well-designed point cloud blocks without additional effort. To evaluate these architectures for dense point cloud analysis, we conduct thorough experiments using S3DIS and ScanNetV2 datasets, where the proposed PointHR outperforms recent state-of-the-art methods without any bells and whistles. The source code is available at \url{https://github.com/haibo-qiu/PointHR}.