Point Cloud Compression with Bits-back Coding

TL;DR

A novel lossless compression method for compressing geometric attributes of point cloud data with bits-back coding that can achieve a competitive compression ratio as conventional deep learning-based approaches, while significantly reducing the overhead cost of storage and/or communicating the compression codec.

Abstract

This paper introduces a novel lossless compression method for compressing geometric attributes of point cloud data with bits-back coding. Our method specializes in using a deep learning-based probabilistic model to estimate the Shannon's entropy of the point cloud information, i.e., geometric attributes of the 3D floating points. Once the entropy of the point cloud dataset is estimated with a convolutional variational autoencoder (CVAE), we use the learned CVAE model to compress the geometric attributes of the point clouds with the bits-back coding technique. The novelty of our method with bits-back coding specializes in utilizing the learned latent variable model of the CVAE to compress the point cloud data. By using bits-back coding, we can capture the potential correlation between the data points, such as similar spatial features like shapes and scattering regions, into the lower-dimensional latent space to further reduce the compression ratio. The main insight of our method is that we can achieve a competitive compression ratio as conventional deep learning-based approaches, while significantly reducing the overhead cost of storage and/or communicating the compression codec, making our approach more applicable in practical scenarios. Throughout comprehensive evaluations, we found that the cost for the overhead is significantly small, compared to the reduction of the compression ratio when compressing large point cloud datasets. Experiment results show that our proposed approach can achieve a compression ratio of 1.56 bit-per-point on average, which is significantly lower than the baseline approach such as Google's Draco with a compression ratio of 1.83 bit-per-point.

Figures (6)

• Figure 1: Overview of a point cloud compression pipeline. Decompression can be done in a reversed order.
• Figure 2: (a) Our entropy estimation approach with the proposed Convolutional Variational Autoencoder (CVAE), (b) detailed architecture of the proposed CVAE, and (c) our bits-back coding approach.
• Figure 3: Encode (compress) and decode (decompress) process of the bits-back coding. The rectangles with dashed lines denote the decrease in the code length of the message, and rectangles with solid lines denote the increase in the code length. The probability values above the rectangles are linked to the derivative in (\ref{['eq:elbo-derivative']}). The rectangles are placed in a stack data structure of ANS with the head of the stack on the left-hand side.
• Figure 4: Visualization of the original point cloud (first column), voxel representation of the point cloud (second column), reconstructed point cloud from the VAE (third column), and decompressed voxel representation (fifth column). The first and second rows illustrate a data sample from ShapeNet's test set in different bit-depth values, i.e., $d=6$ and $d=7$, respectively. The third and fourth rows illustrate a data sample (a bedroom scene) from the SUN RGB-D's test set with bit-depth values $d=6$ and $d=7$, respectively.
• Figure 5: Compression ratios (measured in average bit-per-point) of the methods on the Shapenet dataset. The lower the bit-per-point is, the lower the compression ratio (i.e., the ratio of compressed output sequence length to uncompressed input length) can be achieved.