Efficient Deep Super-Resolution of Voxelized Point Cloud in Geometry Compression

Point cloud compression is an essential task for practical applications using point clouds. Most of the previous approaches rely on octree which involves voxelization in coding itself. Distortions derived from can be reduced without increasing bitrate by postprocessing. In this article, we propose a super-resolution method decoded voxelized as postprocessing step geometry compression. The proposed increases resolution predicting occupancy higher voxels than those used to compress original cloud. For efficient prediction, deep neural network based sparse convolution. It highly even large since applies convolution only nonempty space. predicts occupancies represented continuous values each and estimates binary through thresholding procedure. We design dynamic threshold ensure that at least one all predicted occupied order prevent generation regions with missing points. also introduce prediction address sparsity high-resolution voxels. Experiments outdoor indoor datasets demonstrate effectiveness method.