Netflix's Andrey Norkin Talks Future of AV1
In this interview with Andrey Norkin we covered a range of topics, including the future of Libaom-AV1 in the face of the Alliance for Open Media’s seeming focus on SVT-AV1. By way of background, Norkin’s day job is as a senior research scientist for Netflix, where he contributes to Netflix’s video encoding pipeline, particularly relating to high dynamic range (HDR) and low-bitrate mobile encodes. However, we spoke with him in his role as Co-Chair of the Video Codec Working Group for the Alliance for Open Media.
Jumping right in, regarding the future of libaom-AV1, Andrey explained that though both Libaom-AV1, the AV1 codec in FFmpeg, and SVT-AV1 will be supported going forward, they serve differentfunctions. Specifically, Libaom-AV1 is “the reference software” designed to “cover the whole AV1 standard.” In contrast, SVT-AV1 “is more of a production code base or the code base that is optimized to work ... multi core machines.”
Then the conversation moved into AV1’s playback compatibility. Andrey reviewed AV1’s integration into smart TVs and mobile devices, which has been slower than expected but bolstered by the strong performance of several software decoders, in particular dav1d.
Next, we discussed AV1 and High Dynamic Range video a must for premium content producers like Norkin’s employer Netflix. Norkin first advised that HDR was incorporated into the AV1 spec from day one, and that AOMedia has been working to clarify and document how to deploy HDR 10 and HDR 10+ with AV1. That work is now complete, he reported, so “HDR 10 and HDR 10+ should be supported,” though Dolby Vision is not. Norkin detailed how HDR should work on the playback side, and briefly the type of testing Netflix would perform before starting to distribute AV1-encoded HDR content to AV1 players.
One topic near and dear to Norkin is film grain synthesis, which he contributed to on the development side. As Norkin related, film grain is a very common effect used by film producers, but it complicates encoding because film grain is very close to random noise, which is theoretically uncompressible. The film grain synthesis feature in AV1 removes grain from the original sequence, encodes the video, and sends an estimate of the film grain parameters to the player, which adds the film grain back in an attempt to match the source. The impact is a significant increase in
compression efficiency with minimal impact on quality.
Then we touched on AV1’s current and planned suitability for AR, VR and other applications in the metaverse, and concluded with a brief look ahead at AV2. The conversation was technical and far ranging, we greatly appreciated the opportunity to speak with one of the key contributors to AV1 and to the video quality that most of us watch on Netflix every night.
With the recent launch of version 1.0, SVT-AV1 appears to have caught up with libaom in quality, with very definite performance advantages. Its two-pass rate control is tested and proven. If you're creating an AV1 encoding workflow today that emphasises encoding speed and quality, SVT-AV1 should definitely be on your short list.