3d indoor scene semantic segmentation using 2d semantic segmentation projection

Along with the development of 2D semantic segmentation in the pixel phase of 2D images using deep learning, the technique of detecting objects in 3D space is also emerging. Attempts to detect and split objects in 3D space are currently being actively carried out, but still show lower accuracy compared to 2D pixels. 3D point cloud data has the advantage of providing accurate distance relationship information for a given range of points, but it has irregular and unstructured limitations compared to segmentation within 2D pixel space. In this paper, the method of partitioning indoor point cloud data through 2D semantic segmentation is considered by adopting recently prosed point painting algorithm.