PU-GACNet: Graph Attention Convolution Network for Point Cloud Upsampling
作者：
Highlights：
• We first propose a Graph Attention Convolution (GAC) module as a feature extractor using graph convolution with attention mechanism, which can better establish a joint between the global context aggregation and local context modeling.
• To effectively preserve edge features for new point features generation, we also design an Edge-aware NodeShuffle (ENS) module as a feature expander to consider features of both each point and its neighbors with a shuffle operation.
• Finally, we integrate GAC module and ENS module into a novel point cloud upsampling architecture called PU-GACNet. Results of extensive quantitative and qualitative experiments prove that our method demonstrates superiority on both local and global features extraction and expansion.
摘要
•We first propose a Graph Attention Convolution (GAC) module as a feature extractor using graph convolution with attention mechanism, which can better establish a joint between the global context aggregation and local context modeling.•To effectively preserve edge features for new point features generation, we also design an Edge-aware NodeShuffle (ENS) module as a feature expander to consider features of both each point and its neighbors with a shuffle operation.•Finally, we integrate GAC module and ENS module into a novel point cloud upsampling architecture called PU-GACNet. Results of extensive quantitative and qualitative experiments prove that our method demonstrates superiority on both local and global features extraction and expansion.
论文关键词：Point cloud upsampling,Graph attention convolution,Feature extraction,Edge-aware nodeshuffle,Feature expansion
论文评审过程：Received 18 November 2021, Accepted 28 December 2021, Available online 5 January 2022, Version of Record 12 January 2022.
论文官网地址：https://doi.org/10.1016/j.imavis.2021.104371
