MPEG advances AI coding standards and 12th edition of AVC
MPEG held its 154th meeting, where it advanced several standardization projects, including the 12th edition of AVC and new AES-256 encryption support for ISOBMFF. The meeting also featured demonstrations of MPEG-I immersive audio and progress on AI-related coding standards like Feature Coding for Machines (FCM) and an amendment to Neural Network Coding (NNC), alongside new specifications for point cloud and Gaussian splat coding.
Key Takeaways
- AVC 12th edition reaches Final Draft of International Standard (FDIS) status, adding 11 new SEI messages including AI usage restrictions.
- Feature Coding for Machines (FCM) enters Committee Draft stage to enable split inference applications between edge and cloud devices.
- Encryption of ISOBMFF now supports AES-256 for both CTR and CBC protection mechanisms to improve content security.
- MPEG-I immersive audio demonstrations showcased 6DoF interactive experiences running on contemporary mobile phones and HMDs.
- Geometric-based Point Cloud Coding (G-PCC) was extended to support dense solid objects and Gaussian splat coding.
Why It Matters
The shift toward machine-centric video signals a major transition in the streaming stack, where cameras increasingly feed AI models rather than human viewers. By standardizing feature coding (FCM), MPEG is enabling low-bandwidth, privacy-preserving 'split inference' where compute is distributed between sensors and the cloud. For traditional broadcasters, the updates to AVC ensure that the industry's most widely deployed codec remains relevant via new metadata for AI restrictions and generative video. These standards provide the interoperability layer necessary for the commercial scaling of digital twins, autonomous systems, and 6DoF immersive environments. Watch for the July 2026 launch of the final FCM specification as a catalyst for edge-AI hardware adoption.
Additional Context
The standardization of Feature Coding for Machines (FCM) is a response to the growing dominance of machine-to-machine visual traffic. Per reports from Florida Atlantic University and OP Solutions in late 2025, FCM can reduce required bandwidth by up to 90% compared to traditional pixel-based streaming. This efficiency is critical for autonomous vehicle fleets and drones where network congestion and latency pose safety risks. While traditional codecs like AVC require over 100 Mbps for reliable remote object detection in high-fidelity scenarios, FCM can achieve comparable accuracy at under 10 Mbps by transmitting only essential neural features instead of raw pixels. Simultaneously, the industry is moving to standardize Gaussian Splatting as a more efficient alternative to meshes and point clouds for photorealistic 3D rendering. In February 2026, the Khronos Group announced the KHR_gaussian_splatting extension for glTF 2.0 to ensure cross-platform support. MPEG’s current work on geometry-based Gaussian splat compression (G-PCC) complements these efforts, aiming to eliminate the fragmentation caused by proprietary file formats. According to recent search data, companies like Nokia are leading these contributions to make volumetric video practical for existing mobile consumer hardware. Despite the rise of newer codecs like VVC (H.266), AVC remains the cornerstone of global streaming. Technical briefings from Bitmovin and The Broadcast Bridge in June 2026 note that AVC still handles over 90% of global video traffic. The 12th edition’s focus on Supplemental Enhancement Information (SEI) messages for generative face video and AI restrictions reflects the standard's adaptation to a market increasingly concerned with content authenticity and provenance in the age of deepfakes.
Read full article at mpeg.org
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source