The Alliance for Open Media (AOMedia) will release the AV2 open video codec specification by the end of the year, as an update of the now widely-used AV1 video codec with up to 40% bandwidth reduction at the same level of quality.
In the meantime, Debargha Mukherjee, Google Principal Engineer, has shared a progress report about AV2, first offering a historical perspective with the list of digital video codecs from 1990 to 2030 (expected), then stating AV2 goals, and going through requirements, coding tools, and what’s been achieved so far. You can check the eleven-minute presentation (04:27 – 15:30) embedded at the end of this post, or continue reading for the highlights of the talk.
AV2 will be released very soon, while the H.267 proprietary video codec might only be released around 2030. We haven’t seen a lot of support/traction for H.266, so we’ll have to see if H.267 ever sees the light of day.
Development for AV2 started in 2020 with the goals of achieving a 40% lower bandwidth than AV1 and keeping the video decoder complexity/area in check (not over 2x over AV1) to keep the hardware decoder price competitive. The tools and source code for the AV2 reference implementation are all hosted on GitLab.
Debargha then goes through the highlights of AV2. First, we learn that AV2 is not an AI codec, although it relies on some data-driven tools, so he calls it a “conventional hybrid codec”. He then lists the various advances in intra- and inter-frame compression, transforms, filters, and so on. Lots of technical jargon that’s hard to understand if you’re not familiar with recent advances in image/video processing…
AV2 delivers a “hardware-efficient design”, which probably means AV2 is still quite more complex than AV1 and takes more resources, but they’ve made efforts to minimize increases in bandwidth, RAM, and storage requirements. AV2 significantly improves lossless coding and film-grain synthesis over AV1, and also adds support for Chroma formats (YUV) 4:2:2 and 4:4:4, on top of the existing 4:2:0 format.
The final part of the presentation covers AV2 performance. So did they achieve their goals? The first slide shows the results of objective metrics with automated tools. Debargha explains that the RA row (Random Access) is the most important, and here AV2 shows about 30% lower bitrate required over an improved AV1 implementation. PSNR standard for Peak Signal-to-Noise Ratio and VMAF (Video Multimethod Assessment Fusion) is a video quality assessment algorithm developed by Netflix.
The final slide shows subjective AV2 performance metrics. I did these types of tests for image compression when I was an intern at the University of Bath, and it consisted of getting people in front of a computer with two identical displays to show them photos side-by-side compressed with different parameters, and let them select which one looked best. Google did something similar with AV2 and (modified) AV1 4K UHD videos, and found that on average (BD-rate) column AV2 needed 38% fewer bits than the improved AV1 at the same perceived quality. For one sample, (DrivingPOV3), the BD-rate value even drops to -50.63%.
AOMedia’s “The Future of Innovation is Open” webinar’s video is embedded below. The full video is one hour and 22 minutes long, but the AV2 presentation is only 11 minutes long.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress. We also use affiliate links in articles to earn commissions if you make a purchase after clicking on those links.
