deep learning technique has yielded significant improvements in point cloud
completion with the aim of completing missing object shapes from partial
inputs. However, most existing methods fail to recover realistic structures due
to over-smoothing of fine-grained details. In this paper,