Deep Geometry Post-Processing for Decompressed Point Clouds

TL;DR

This paper introduces a learning-based post-processing method that enhances decompressed point clouds by predicting occupancy probabilities with a 3D convolutional network, significantly improving quality across diverse distortions.

Contribution

It presents a novel single-model approach using multi-scale 3D CNNs for geometry refinement of decompressed point clouds with various distortions.

Findings

Achieves an average of 9.30dB BDPSNR gain.

Effectively handles diverse distortions with a single model.

Significantly improves point cloud quality after compression.

Abstract

Point cloud compression plays a crucial role in reducing the huge cost of data storage and transmission. However, distortions can be introduced into the decompressed point clouds due to quantization. In this paper, we propose a novel learning-based post-processing method to enhance the decompressed point clouds. Specifically, a voxelized point cloud is first divided into small cubes. Then, a 3D convolutional network is proposed to predict the occupancy probability for each location of a cube. We leverage both local and global contexts by generating multi-scale probabilities. These probabilities are progressively summed to predict the results in a coarse-to-fine manner. Finally, we obtain the geometry-refined point clouds based on the predicted probabilities. Different from previous methods, we deal with decompressed point clouds with huge variety of distortions using a single model. Experimental results show that the proposed method can significantly improve the quality of the decompressed point clouds, achieving 9.30dB BDPSNR gain on three representative datasets on average.