I couldn't find this presentation on Vimeo. This is really a talk about Video Compression, not video codecs in general, but there's a lot more to say about compression than lossless codecs.

Derek Buitenhuis is the senior principal video engineer at Vimeo, although he's been deeply involved with ffmpeg and VideoLAN (among other open source projects).

Derek talks about how the only stakeholders really interested in next-generation video codecs like AV1 or VVC (and even last-gen like VP8/VP9 and HEVC) are megacorporations or people actually involved in developing the standards themselves. Derek's argument is that the bandwidth savings you get with extra efficiency aren't that great compared with the other costs you have to pay; H.264 is good enough for most people and businesses today.

Derek asserts that compute is a bigger issue today than bandwidth; and a lot of these newer codecs are not saving much on computational complexity. If anything, they are actually far more computationally complex. This is true for AV1 especially, as not a lot of hardware encoders/decoders are out for it yet. And the most commonly-used encoder for HEVC, for example, is not that great because FFmpeg's developers are not interested in working on it given the dicey legal situation.

He also talks about how no one is really innovating on as large a scale as they were in the early 2000s, and that a lot of deep knowledge about how or why codecs work a certain way can't easily be obtained by engineers new to video codecs. Part of this is poor documentation, and part of it is because newer codecs are being built off the back of old stuff. No one except the people who were around in the beginning really knows why certain stuff is done a certain way.

So, is that it? Are we done? Do we just ride out the last few years of baseline patents for H.264 and standardize on that? Will the web finally have a standard video codec recommendation? I mean, a lot of the audio industry hasn't moved on from mp3. If they have, it's only to AAC...

Personally, I would prefer further innovation but not at the cost of more compute, and further innovation in the direction of AV1 and EVC, not VVC. I'm sure that's an uncontroversial opinion.

Further discussion on the Lightworks Forum (I'm parttimeNLE): https://forum.lwks.com/threads/vimeo-engineer—are-video-codecs-done-h-264-is-the-jpeg-of-video.250235/

September 14, 2023 Update: A Lightworks staff member shared a link to Meta's adoption of AV1 with Facebook Reels, and it was super interesting. It serves as a very good counter-argument to Derek's presentation: https://engineering.fb.com/2023/02/21/video-engineering/av1-codec-facebook-instagram-reels/

Comments are closed