Standards Development

Scope

This project will define bitstream syntax and semantics of encoded point clouds, and decoding procedure to reconstruct point clouds.

Purpose

Dynamic point clouds are sequences of 3D point cloud frames that represent the surfaces of moving objects or evolving scenes over time. This data format is increasingly used in applications such as LiDAR sensing, autonomous driving, virtual and augmented reality (VR/AR), and immersive communications.

Each point in a point cloud is defined by its 3D coordinates, which capture the geometry of the scene, and may carry additional attributes depending on the application context—for instance, reflectance values for LiDAR-based systems or RGB color for visual rendering. These rich data structures support tasks such as 3D reconstruction, object detection, and semantic understanding in machine perception pipelines. However, dynamic point cloud sequences are characterized by extremely high data volumes, which pose significant challenges for transmission, storage, and real-time processing. There is thus a pressing need for efficient, scalable, and interoperable compression technologies.

The proposed new work item addresses this need by introducing the first MPEG standard dedicated to the compression of dynamic point clouds using AI-based technologies. The aim is to define a neural network–driven framework for geometry compression that can adapt to a wide spectrum of point cloud densities—from sparse formats like LiDAR to dense representations used in immersive visual applications. While existing MPEG standards such as G-PCC and V-PCC rely on handcrafted coding tools inspired by traditional video and geometry compression, AI-based approaches offer a paradigm shift by learning compact representations directly from data, enabling better adaptation to content diversity, improved compression performance, and new capabilities such as semantic-aware coding. By leveraging data-driven modeling and inference capabilities, this AI-based approach is expected to surpass traditional signal processing techniques in terms of compression performance and flexibility. Furthermore, the framework may be extended to support the compression of various point attributes, enabling its use across an even broader range of industrial and research applications.

This new standard has the potential to become a key enabler for next-generation 3D media systems, robotics, and intelligent sensing platforms.