Integration of Turing Codec (HEVC encoder)

[Kenneth Peard]

At the moment the only software HEVC encoder ffmpeg uses is libx265, and i don't know if you want to keep it that way, but in 2016 the BBC Research and Development institute release their "Turing Codec" which is a HEVC encoder licensed under the GPL v2. Would it be possible to add integration to ffmpeg?

Github repo is found here:
​https://github.com/bbc/turingcodec

[Gyan]

That encoder was submitted and rejected. See ​http://www.ffmpeg.org/pipermail/ffmpeg-devel/2017-December/222341.html

[Carl Eugen Hoyos]

Replying to fantesy:

At the moment the only software HEVC encoder ffmpeg uses is libx265, and i don't know if you want to keep it that way, but in 2016 the BBC Research and Development institute release their "Turing Codec" which is a HEVC encoder licensed under the GPL v2. Would it be possible to add integration to ffmpeg?

Why do you believe this would be a good idea?
(Your answer may be able to invalidate the reasons for not accepting the patch.)

[Kenneth Peard]

ok, my apologies I didn't know that it had already been submitted and rejected. I wasn't suggesting it based on performance or capabilities, more of giving people a different option and it was the only other HEVC encoder I was aware of. However if you have your reasons for not including it, I can accept that and I will close this ticket.

And again i would like to apologies, I wasn't aware that they had been bugging you guys about integrating it, I don't keep as up to date on the mailing list as I would like.

[jkqxz]

H.265 encoders currently available in ffmpeg:

• x265 (software via libx265)
• Kvazaar (software via libkvazaar)
• Intel (software and hardware encoders via libmfx, or hardware-only via VAAPI)
• AMD (hardware via AMF or VAAPI)
• Nvidia (hardware via NVENC)

I think Turing would probably be accepted if there were any particular argument to do so, but currently it just looks like a closed research project:

• It's pretty much stated that the API isn't stable, hence the suggested patch avoiding any common options and using the string argument stuff to push any incompatibility onto users.
• The open-source project looks like an unsupported wall-throw - e.g. <​https://github.com/bbc/turingcodec/commit/09a838c9feeb43ac410391e300b8b53ff660c721>.
• There was little attempt to show that it's actually good for anything - no visible users, no useful benchmarks.