Overview
ISO/IEC 21794-6:2025 - "Information technology - Plenoptic image coding system (JPEG Pleno) - Part 6: Learning‑based point cloud coding" specifies a learning‑based framework for point cloud compression within the JPEG Pleno family. The standard defines a single‑stream coded codestream format, encoding tools, and extensions to the JPEG Pleno File Format and metadata descriptors that are specific to point cloud modalities. It targets interactive human visualization and machine tasks, aims for competitive compression efficiency, and has the goal of supporting a royalty‑free baseline.
Key topics and technical requirements
- Learning‑based compression: Uses machine learning to learn compressed domain representations from training data (trained models and parameters are addressed).
- Codestream and file format: Specifies a coded codestream format for storage of point clouds and extensions to the JPEG Pleno File Format.
- Encoder / decoder requirements: Functional descriptions and mandatory requirements for encoders and decoders are defined to ensure interoperability.
- Data representations: Covers point cloud geometry, sparse/dense tensors, latent tensors, and attribute handling (e.g., colour).
- Neural network constructs: Defines neural network layers, modules and model concepts (sparse convolution, transposed/generative sparse convolutions, quantized layers).
- Entropy and binarization: Includes entropy coding and binarization mechanisms relevant to compact codestream representation.
- Metadata and random access: Extensions for metadata descriptors and support for interactive access and advanced functionality (editing, random access, rights protection).
- Normative annexes: Detailed normative material includes file format, geometry codestream syntax, deep‑learning decoding, colour decoding, synthesis transforms, hyper decoders, entropy decoding, block merging, up‑sampling and super‑resolution.
Applications and who should use it
- Codec developers and implementers building point cloud encoders/decoders for visualization platforms.
- 3D visualization and AR/VR platforms that require efficient, interactive streaming and storage of point clouds.
- Computer vision and machine learning practitioners who need compressed representations optimized for downstream 3D processing or recognition tasks.
- Cloud storage and streaming services delivering volumetric content with metadata and random access.
- Standards bodies and system integrators aligning implementations with JPEG Pleno and related volumetric media standards.
Related standards
Keywords: ISO/IEC 21794-6:2025, JPEG Pleno, learning-based point cloud coding, point cloud compression, 3D point cloud, sparse tensor, neural network codec, codestream, royalty-free baseline.