IPDAE: Improved Patch-Based Deep Autoencoder for Lossy Point Cloud Geometry Compression

TL;DR

This paper introduces IPDAE, an enhanced patch-based deep autoencoder for lossy point cloud compression, combining learnable context modeling, octree sampling, and adversarial training to improve efficiency and quality.

Contribution

The paper presents novel improvements to patch-based point cloud compression, including a learnable context model, octree sampling, and adversarial training, achieving better rate-distortion performance.

Findings

Outperforms state-of-the-art in rate-distortion metrics.

Maintains short compression time with high reconstruction quality.

Effective on both sparse and large-scale point clouds.

Abstract

Point cloud is a crucial representation of 3D contents, which has been widely used in many areas such as virtual reality, mixed reality, autonomous driving, etc. With the boost of the number of points in the data, how to efficiently compress point cloud becomes a challenging problem. In this paper, we propose a set of significant improvements to patch-based point cloud compression, i.e., a learnable context model for entropy coding, octree coding for sampling centroid points, and an integrated compression and training process. In addition, we propose an adversarial network to improve the uniformity of points during reconstruction. Our experiments show that the improved patch-based autoencoder outperforms the state-of-the-art in terms of rate-distortion performance, on both sparse and large-scale point clouds. More importantly, our method can maintain a short compression time while ensuring the reconstruction quality.