Transcription of PCT: Point Cloud Transformer - arXiv
{{id}} {{{paragraph}}}
Example: bankruptcy
PCT: Point Cloud Transformer - arXiv
PCT: Point Cloud TransformerMeng-Hao GuoTsinghua CaiTsinghua LiuTsinghua MuTsinghua R. MartinCardiff HuTsinghua irregular domain and lack of ordering make it chal-lenging to design deep neural networks for Point Cloud pro-cessing. This paper presents a novel framework namedPoint Cloud Transformer (PCT)for Point Cloud is based on Transformer , which achieves huge successin natural language processing and displays great potentialin image processing. It is inherently permutation invariantfor processing a sequence of points , making it well-suitedfor Point Cloud learning. To better capture local contextwithin the Point Cloud , we enhance input embedding withthe support of farthest Point sampling and nearest neighborsearch. Extensive experiments demonstrate that the PCTachieves the state-of-the-art performance on shape classifi-cation, part segmentation, semantic segmentation and nor-mal estimation IntroductionExtracting semantics directly from a Point Cloud is anurgent requirement in some applications such as robotics,autonomous driving, augmented reality, etc.
PointNet [21] pioneered point cloud learning. Subse-quently, Qi et al. proposed PointNet++ [22], which uses query ball grouping and hierarchical PointNet to capture lo-cal structures. Several subsequent works considered how to define convolution operations on point clouds. One main approach is to convert a point cloud into a regular voxel
Loading..
Tags:
Information
Domain:
Source:
Link to this page:
Please notify us if you found a problem with this document: