A testbed of the AI in Media Services project.
3GPP covers AI/ML for 5G media in two places. SA2 (system architecture) defines the Network Data Analytics Function (NWDAF), which collects data from network functions in the 5G core and produces analytics and predictions that other functions can consume. SA4 (media codecs and delivery) studies AI and ML in 5G media services, in TR 26.927 and TR 26.847.
5G-MAG's AI/ML Evaluation Framework is the repository cited by TR 26.847 for the AI4Media evaluations, with the scenarios collected for the FS_AI4Media study. It is described in the repository README and covered below.
This project is based on technology developed by:
The Problem It Solves
A claim that an AI/ML model improves media quality, ABR selection or traffic classification is only useful if it can be checked against a common baseline; without one, every vendor's benchmark is measured on its own datasets and its own terms, and results can't be compared. The Evaluation Framework provides that common basis: shared benchmarks and datasets that let different AI/ML approaches be measured the same way, feeding directly into 3GPP SA4's own evaluation study (TR 26.847) instead of leaving that evidence to be assembled separately by each contributor.
The project at a glance
What it is, how it works, and where the Reference Tools stand, in one deck you can share.
Download the deck (AI/ML Evaluation Framework)From idea to running code
Read it, check it against the standard, then run it.
Specifications
The specifications this technology is built on.
Software Accelerator
The open-source software that implements it.
Open Source, Built Together
Repositories of This Project
The real, running software for this technology: what each repository is, what it implements, its latest release and where to find it.
Early Access: available to members first
Join the Effort
Help shape what connected media looks like next and scale your projects with the industry.