Recently, the deep-learning-based PatchMatch method has been rapidly developed in 3D reconstruction, based on which boundary regions are filled with other parts that most closely match edge parts, but limited data hinder generalization of to unknown settings. If various large-scale datasets generated, process would require considerable time and resources when performing neighborhood point-match...