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.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to efficiently compress the geometric properties of point cloud data while reducing the additional overhead generated during the compression process**. Specifically, the author proposes a lossless compression method based on bits - back coding for compressing the geometric properties of point cloud data. This method estimates the Shannon entropy of point cloud information through the Convolutional Variational Autoencoder (CVAE) in deep learning, and uses the learned CVAE model and bits - back coding technology to compress point cloud data. ### Main problems and solutions 1. **Efficiently compress point cloud data**: - Point cloud data has a large number of geometric properties (such as 3D coordinates), which makes its storage and transmission very large. For example, a LiDAR sensor can generate 8.4 billion points per day. - Traditional compression methods are less efficient when dealing with large - scale point cloud data and need to store or transmit marginal probabilities, which will bring additional overhead. 2. **Reduce the additional overhead during the compression process**: - Current entropy estimation models based on deep learning assume that the decoder can access the learned marginal probabilities, but in practical applications, this assumption is often unrealistic because the encoder and decoder need to store or communicate these probabilities, which may exceed the size of the compressed data itself. - To this end, the author proposes a method using bits - back coding, which avoids the need for marginal probabilities, thereby significantly reducing the additional overhead. ### Method overview - **Convolutional Variational Autoencoder (CVAE)**: It is used to estimate the Shannon entropy of point cloud data. CVAE processes high - dimensional voxel data through 3D convolutional layers to capture and learn the spatial information of point cloud. - **Bits - back Coding**: It uses the learned latent variable model for compression without the need to access marginal probabilities. This method can capture the potential correlations between point clouds and further reduce the compression ratio. ### Experimental results The experimental results show that the proposed bits - back coding method performs excellently when compressing large - scale point cloud data, with an average compression ratio of 1.56 bit - per - point, which is better than Google's Draco library (1.83 bit - per - point). In addition, this method shows significant advantages in both compression ratio and additional overhead, and is suitable for efficient point cloud compression in practical application scenarios. ### Formula summary - Shannon entropy formula: \[ H(x_m)=\sum_{i = 0}^{1}P(x_m = i)\log\frac{1}{P(x_m = i)} \] - Chain rule of conditional probability: \[ P_{\theta}(x)=\prod_{m = 1}^{M}P_{\theta}(x_m|x_1,x_2,\ldots,x_{m - 1}) \] - Free energy of bits - back coding: \[ L_{\theta;\phi}=E_{Q_{\phi}(z|x)}[\log P_{\theta}(x|z)+\log P(z)-\log Q_{\phi}(z|x)] \] Through these methods, the author has successfully solved the key problems in point cloud data compression and achieved an efficient and low - overhead compression scheme.

Point Cloud Compression with Bits-back Coding

3QNet: 3D Point Cloud Geometry Quantization Compression Network

3QNet

Lossless Point Cloud Geometry and Attribute Compression Using a Learned Conditional Probability Model

DeepCompress: Efficient Point Cloud Geometry Compression

End-to-end learned Lossy Dynamic Point Cloud Attribute Compression

3D Point Cloud Attribute Compression Using Geometry-Guided Sparse Representation

Lossy Geometry Compression Of 3d Point Cloud Data Via An Adaptive Octree-Guided Network

Learned Point Cloud Geometry Compression

Learning Convolutional Transforms for Lossy Point Cloud Geometry Compression

Lossy Point Cloud Geometry Compression Via End-to-End Learning

3D Point Cloud Geometry Compression on Deep Learning

Deep AutoEncoder-based Lossy Geometry Compression for Point Clouds

Point Cloud Compression with Implicit Neural Representations: A Unified Framework

Virtual Structured-Light Three-Dimensional Point Cloud Compression with Geometric Reshaping

Density-preserving Deep Point Cloud Compression

Multiscale Point Cloud Geometry Compression

Point cloud geometry compression with sparse cascaded residuals and sparse attention

Point AE-DCGAN: A Deep Learning Model for 3D Point Cloud Lossy Geometry Compression.

Point Cloud Geometry Compression Based on Multi-Layer Residual Structure

Survey on Deep Learning-Based Point Cloud Compression