Called the Video Coding for Machines (VCM) Ad-hoc Group (AhG), the initial focus will be mainly compression efficiency, taking account of the fact that ability to recognize objects quickly and accurately is the goal, rather than enjoyment of the experience. The aim is therefore to seek compression performance greater than that achieved by current or forthcoming codecs for transmission of content to humans, such as Versatile Video Coding (VVC).

This comes when Cisco among others have been predicting that machine-to-machine applications will generate the fastest growth in internet video traffic over the next few years. This means that efficient compression of video data for machine use will be important for competitiveness and also for ensuring there is sufficient capacity for all applications and services, including those streaming to humans.

While the aim with conventional video coding is to compress and then reconstruct whole frames with a view to achieving the most enjoyable perception possible at the target resolution, for machines it is to preserve just critical information. But machines will vary in their requirements and so the focus of research now is to apply AI techniques to adapt compression for specific use cases, with the advantage being that success is somewhat easier to define via testing, or at any rate more direct to establish in the machine case. If the machine can perform its allotted tasks accurately enough, then video will be deemed to have been reconstructed satisfactorily. The objective would be to achieve the lowest bit rate at which performance or safety targets are met, presumably leaving some headroom.

The idea of a new codec called VCM was proposed earlier in August 2019 by China Telecom in conjunction with Gyrfalcon Technology, a developer of AI accelerators. The need had just been recognized after over 40 years of video compression history led by MPEG. The stated aim was to develop vision chips for a variety of sectors in the burgeoning Internet of Things (IoT) arena.