Learning Convolutional Transforms for Lossy Point Cloud Geometry Compression

Abstract

Efficient point cloud compression is fundamental to enable the deployment of virtual and mixed reality applications, since the number of points to code can range in the order of millions. In this paper, we present a novel data-driven geometry compression method for static point clouds based on learned convolutional transforms and uniform quantization. We perform joint optimization of both rate and distortion using a trade-off parameter. In addition, we cast the decoding process as a binary classification of the point cloud occupancy map. Our method outperforms the MPEG reference solution in terms of rate-distortion on the Microsoft Voxelized Upper Bodies dataset with 51.5% BDBR savings on average. Moreover, while octree-based methods face exponential diminution of the number of points at low bitrates, our method still produces high resolution outputs even at low bitrates. Code and supplementary material are available at https: //github.com/mauriceqch/pcc_geo_cnn.

Learning-Based Lossless Compression of 3D Point Cloud Geometry

“…Inspired by the success in learning-based image compression, deep learning has been recently adopted in point cloud coding methods [12][13][14][15][16]. The proposed methods in [15,16] encode each 64 × 64 × 64 sub-block of PC using a 3D convolutional auto-encoder. In contrast, in this paper we losslessly encode the voxels by directly learning the distribution of each voxel from its 3D context.…”

Improved Deep Point Cloud Geometry Compression

“…A high α value makes marking occupied voxels as empty more costly than marking empty voxels as occupied and results in denser reconstructions. Originally, we picked the same α value (0.90) as in [4]. This was motivated by the fact that point clouds are often comprised of more than 95% of empty space.…”

“…We then present an ablation study identifying key performance factors for DPCC. In particular, we start from a baseline DPCC model [4] and we consider the following improvements:…”

Pc-Mcu

“…Additional methods to improve the Quality of Experience (QoE) are planned to be included in the next stages of the development. One interesting method is the one by Quach et al [16], based on adjusting the size of the points. Anyway subjective studies need to be conducted to determine the most appropriate strategies.…”