Abstract
The size of the input receptive field is one of the most critical aspects in the semantic segmentation of the point cloud, yet it is one of the most overlooked parameters. This paper presents the multiple-input receptive field processing semantic segmentation network MRNet. The fundamental philosophy of our design is to overcome the size of the input receptive field dilemma. In particular, the input receptive field's size significantly impacts the performance of different sizes of objects. To overcome this, we introduce a parallel processing network with connection modules between the parallel streams. Our ablation studies show the effectiveness of implemented modules. Also, we set the new state-of-art performance on the large-scale point cloud dataset SensatUrban.
Abstract (translated)
输入接收场的大小是点云语义分割中最为重要的方面之一,但却是最容易被忽视的参数之一。本文介绍了多输入接收场处理语义分割网络MRNet。我们的设计哲学是克服输入接收场的大小难题。特别是,输入接收场的大小对不同大小的物体的性能有着显著的影响。为了克服这个问题,我们引入了一个并行处理网络,其中在并行流之间的连接模块。我们的研究结果表明所实现的模块的有效性。此外,我们还在大规模点云数据集 SensatUrban 上实现了新的先进性能。
URL
https://arxiv.org/abs/2301.12972