What would you like to ask Multicoreware? (IBC Meeting)

[FranceBB]

Hi there, everyone.
As some of you know I'm gonna meet with the Multicoreware guys on Monday at IBC (https://forum.doom9.org/showthread.php?t=184144) and I'd like to collect some questions from everyone here at Doom9 so that I can bring those to them.
As things currently stands, whenever we use default presets like --preset slower or --preset veryslow etc they won't make use of any new added feature from version 3.5, 3.4, 3.3, 3.2, 3.1 as the last time they were modified was on January 23rd, 2019 when version 3.0 was released and I do think that it's about time to review and update them. Another thing would be to modify tunes like --tune grain.
Can we, as a community, come up with a strong proposal about what we think should be included by default in the presets and isn't included right now?
Also, if you had the chance, what would you like to ask them?

Questions gathered so far:

1) Over the last few months the state of the x265 project doesn't seem to have been progressing that much, especially this year, with the number of merged commits being just as little as 18 (https://bitbucket.org/multicoreware/x265_git/commits/). Is this because all the devs are now focusing on x266?

2) When the x265 development went live in 2013, you used to have developers interacting directly with users of Doom9, getting feedback from both professionals and home users who are really passionate about it. This was a really nice way to improve things 'cause we as a community used to get the input directly from you and you as devs used to get constant feedbacks from us. Unfortunately, the x265_Project account has been closed and deleted and now we feel like there's a gap between end users and devs. Are there any plans to bring it back? Why was it closed? Is there gonna be one for x266 too?

3) Given the not-so-recent departure from Multicoreware of the head of engineering for video products Pradeep Ramachandran, who has taken his place? Is this affecting the release of x266 or do you think things are going smoothly and will be released just as fine? Speaking of releases, as things stands the repository is still private, do you have a date (even a wild guess/estimate) on when it's gonna be public? Would it be possible to have a test build?

4) Recently Sony has introduced their new standard, so they moved from XAVC Intra Class based on H.264 with all intra and other constraints to their new HS standard based on H.265. Given that in x264 we have the dedicated options that trigger all the available settings required, like x264 --avcintra 300 --avcintra-flavor sony, are we gonna see them included into x265 as well for their new HS-H.265 based standard?



Please let me know if you have anything to ask them and I'll add it to the questions.
The meeting will be on Monday, September 12th.

Cheers,
Frank

[excellentswordfight]

I don’t have any direct question, I'm mostly interested in the official status the x265 development is as it seems from the outside be on life-support, and what the status of video-encoding development in general looks like at Multicoreware, and if they are going to continue, or rather go back to a more open-development (like you mentioned with more open discussions with end users).

I’m not sure if that is the right forum for specific x265 tweaking, I would rather have them to come back to doom9 for example were we could have input on presets etc. But while on the topic my 2 cents would be:

Given the huge speed increase rskip 2 can provide while not having much of an negative impact on the image it should be considered to be included in the presets (maybe all the way down to slower, 1 for veryslow and 0 for placebo). Selective-sao 2 also seems like more balanced than sao and could be included from medium down to placebo. I have also always thought, especially now when high core count systems have become so common, that ctu and me-range should by default change based on input resolution, using 32 as default for 720p and below.

I would keep tune grain as rather niche, focusing on visually lossless grain retention for enocdes at high bitrates, e.g. uhd-bluray authoring (still might need some changes though), while adding a new tune film for more general use (maybe something like --aq-mode 1 --no-sao --deblock -1,-1 --psy-rd 2.5 --psy-rdoq 4).

And while we are at it, since keyint, min-keyint & lookahead all have values based on framrate, why not make the values dynamic. keyint=fps*10 (max 300), min-keyint=fps*1, lookahead=fps (*1-4 based on preset, max 120).

[HD MOVIE SOURCE]

1. What are the biggest outliers for bugs using the x265 encoder that need fixing?
2. Are they any updates to x265 that could significantly improve encoding quality or even encoding speed?

[vxzms]

Is there any new news now?

[FranceBB]

Yes, so... we met and had a coffee together and they're very very very nice guys, like, honestly, they even insisted to pay for my coffee.

I met with the new person in charge of the encoding division they have and rest assured that I have good, positive news.
The reason why there haven't been many changes recently in x265 is in fact because they're focusing their efforts on x266 and it took a lot of effort to have it done properly, but now they're very close to have a 1.0 non public test build that they're gonna use for internal tests.
If everything goes well, this internal non public test build should be ready by december for internal testing and in January 2023 it will be available to paying partners like us (yes, Sky is one of the paying Multicoreware partners as we're sustaining the development and it would be nice if many more companies that actively use x26x actually joined and did the same).
We also talked about the x265_Project account and the fact that they lost touch with the open source community and they sort of knew and were familiar with Doom9, so it's very much possible that they'll be back on Doom9, but even if they won't, I'm making sure to be included in the occasional partners-meeting so that I can gather info from the community and then come back to them.
An example is the default presets refactoring and the fact that it hasn't been changed since 2019.
When we talked about this they were very open to the idea, however this MUST come after gathering data, tests and putting together a strong proposal from our side.
The reason is that it would be very time consuming for their devs to read all the discussions here on Doom9 every time, so the way it should work is that we, together, as a community, make tests, vote changing parameters etc and come up with a proposal in which we show all the tests we've done and why we want to turn on a particular option or not and how it does indeed influence the final perceived quality and then me (or other partners) will present it to them and they'll review it and eventually accept it.

[vxzms]

Oh, thanks. x266 coming out soon sounds great, although it is internal.

In fact, I’ve had a question for a long time about how Multicoreware views the x26x, high quality video compression, broadcast, or web streaming? The team’s development in recent years I feel is more towards the last one.

[RanmaCanada]

this is great information. Now let's hope that x266 isn't as fragmented as AV1 has become, and is no where near as difficult to use.

[ReinerSchweinlin]

Yes, so... we met and had a coffee together and they're very very very nice guys, like, honestly, they even insisted to pay for my coffee.

I met with the new person in charge of the encoding division they have and rest assured that I have good, positive news.
The reason why there haven't been many changes recently in x265 is in fact because they're focusing their efforts on x266 and it took a lot of effort to have it done properly, but now they're very close to have a 1.0 non public test build that they're gonna use for internal tests.
If everything goes well, this internal non public test build should be ready by december for internal testing and in January 2023 it will be available to paying partners like us (yes, Sky is one of the paying Multicoreware partners as we're sustaining the development and it would be nice if many more companies that actively use x26x actually joined and did the same).
We also talked about the x265_Project account and the fact that they lost touch with the open source community and they sort of knew and were familiar with Doom9, so it's very much possible that they'll be back on Doom9, but even if they won't, I'm making sure to be included in the occasional partners-meeting so that I can gather info from the community and then come back to them.
An example is the default presets refactoring and the fact that it hasn't been changed since 2019.
When we talked about this they were very open to the idea, however this MUST come after gathering data, tests and putting together a strong proposal from our side.
The reason is that it would be very time consuming for their devs to read all the discussions here on Doom9 every time, so the way it should work is that we, together, as a community, make tests, vote changing parameters etc and come up with a proposal in which we show all the tests we've done and why we want to turn on a particular option or not and how it does indeed influence the final perceived quality and then me (or other partners) will present it to them and they'll review it and eventually accept it.
Thanx for making the effort and aiding in bringing the community wishes to them...

What kind of coffee did you have?

[benwaggoner]

this is great information. Now let's hope that x266 isn't as fragmented as AV1 has become, and is no where near as difficult to use.
Do you mean you hope VVC doesn't become as fragmented? VVC and AV1 are bitstream formats. x266 would be one of multiple different encoder implementations for VVC.

[FranceBB]

Ben! There were too many people and we couldn't meet!! :(
I really expected to see you at the IMF User Group 'cause people from all broadcasting companies and streaming services of the world got there, but instead there were two of your colleagues from Amazon, John Bilt and Raymond Yeung... :(
That's really a shame! :(

[benwaggoner]

Ben! There were too many people and we couldn't meet!! :(
I really expected to see you at the IMF User Group 'cause people from all broadcasting companies and streaming services of the world got there, but instead there were two of your colleagues from Amazon, John Bilt and Raymond Yeung... :(
That's really a shame! :(
Yeah, IBC is a crazy show. My schedule was packed all four days and I still only talked to half the people I'd hoped to.

[hajj_3]

@FranceBB 'Sky Stream' boxes were announced yesterday in the uk. Do those boxes support VP9, AV1 or VVC decoding?

[FranceBB]

'Sky Stream' boxes were announced yesterday in the uk. Do those boxes support VP9, AV1 or VVC decoding?

I have no idea.
Aside from the fact that I work in the Italian branch and not in the British one, the end-user stuff is handled by a department called "Technology" while I work in a totally different department called Encoding Department.
Essentially my "users" are not end-user/customers, but rather the technology guys who are gonna take the flattened high bitrate mezzanine file I created for them and they're gonna re-encode it for the end-user.
In other words, they're the ones who deal with the boxes directly, create the software that runs on them, manage all the end-user renditions in different codecs etc.
There's also a different department called Toc who distributes bitrate according to what is available etc, but again, they deal with end-users, I don't, I just create the files that they're then going to use to create the end-user files/streams.

I also gotta admit that I totally lack any knowledge about streaming as I'm pretty much involved in the linear TV channels that air through satellite, which take the files played locally on a video server and whose SDI signal is then re-encoded by a live hardware encoder, multiplexed and sent on air as .ts, so although I have a wide understanding about how linear supply chains work, I have no knowledge at all about the streaming stuff.

If this was an Italian satellite-based linear channels box, it would have had at the very least MPEG-2 support (for SD), H.264 support (for FULL HD), H.265 support (for UHD) as far as video is concerned, while for audio it would have had MP2 Stereo support for stereo and AC3 support for 5.1.
I have no idea what they do in UK, but I seem to recall from the back of my head that they use E-AC3 for both normal 5.1 and Atmos and probably something else for stereo.
Again, this was only for satellite feeds, so it might be entirely different for streaming and perhaps they use totally different codecs, I don't know.

[hajj_3]

I have no idea.
Aside from the fact that I work in the Italian branch and not in the British one, the end-user stuff is handled by a department called "Technology" while I work in a totally different department called Encoding Department.
Essentially my "users" are not end-user/customers, but rather the technology guys who are gonna take the flattened high bitrate mezzanine file I created for them and they're gonna re-encode it for the end-user.
In other words, they're the ones who deal with the boxes directly, create the software that runs on them, manage all the end-user renditions in different codecs etc.
There's also a different department called Toc who distributes bitrate according to what is available etc, but again, they deal with end-users, I don't, I just create the files that they're then going to use to create the end-user files/streams.

I also gotta admit that I totally lack any knowledge about streaming as I'm pretty much involved in the linear TV channels that air through satellite, which take the files played locally on a video server and whose SDI signal is then re-encoded by a live hardware encoder, multiplexed and sent on air as .ts, so although I have a wide understanding about how linear supply chains work, I have no knowledge at all about the streaming stuff.

If this was an Italian satellite-based linear channels box, it would have had at the very least MPEG-2 support (for SD), H.264 support (for FULL HD), H.265 support (for UHD) as far as video is concerned, while for audio it would have had MP2 Stereo support for stereo and AC3 support for 5.1.
I have no idea what they do in UK, but I seem to recall from the back of my head that they use E-AC3 for both normal 5.1 and Atmos and probably something else for stereo.
Again, this was only for satellite feeds, so it might be entirely different for streaming and perhaps they use totally different codecs, I don't know.

There's a good chance sky will release these boxes in all countries they operate as they will be very cheap to make and no satellite installer to pay so you might want to investigate. VVC support would be helpful for those with slow broadband connections.

[vitaloque]

I'd ask set of predefined encoding profiles based on source and target. E.g.
1) BD Rip, 1080, film, light grain
2) BD Rip, 1080, film, heavy grain
3) Same as 1-2 but for anime
4) BD Rip, 2160 to 1080, HDR10, film, light grain
5) BD Rip, 2160 to 1080, HDR10, film, heavy grain
6) Same as 4-5 but keep the 2160p resolution
7) Same as 4-6 but anime

And for all above 2 options: good quality and high quality. Meaning I'm ok losing some details as long as bitrate doesn't exceed ~9000kbps or I want to keep as much details as it is noticable and file size doesn't matter.

Guess we all need that for a very long time.

[rwill]

I'd ask set of predefined encoding profiles based on source and target. E.g.
1) BD Rip, 1080, film, light grain
<more BD Rip removed>


I don't think Multicoreware wants to target the piracy market with their products.

[FranceBB]

I don't think Multicoreware wants to target the piracy market with their products.

Yeah...

@vitaloque... we're all professionals working in the field here, so there's no room for piracy.

[excellentswordfight]

Yeah...

@vitaloque... we're all professionals working in the field here, so there's no room for piracy.
Yeah right... Although there seems to be a rather healthy amount of professionals active here.

But yeah, the whole issue is that a use case on these forums and for consumers (and pirates) in general (re-compressing distributed media as far as possible without visible quality loss without specific bandwidth/size constraints) is not the same use case the people Multicoreware are developing this encoder for. An 1080p encode looking good at 3Mbps has much more commercial value then one looking excellent at 10Mbps. And a lot of the custom tweaks done to improve quality at CRF <18 will hurt quality at most streaming bitrates.

And given the compute-requirement of modern codecs, streaming, and the lower cost in private storage these medium-sized-private-encode-scenarios are getting more and more niche anyway.

That said, I still think they just should have thrown in a preset comparable to x264s tune film, they could´ve just thrown in no-sao, deblock -1:-1 and a minor psy tweak and called it a day. If it also defaulted to preset slow, that would've alteast removed a ton of posts online :)

And @vitaloque, eitherway I think those things are much more appropriate to implement in a front-end progam.

[benwaggoner]

Yeah. Pretty much the only time I ever use x265 with non-visually lossless content is if I'm reencoding some family video to meet some file sharing maximum file size requirement or something. 99.9% of my x265 encodes are of ProRes, .y4m, J2K, etc sources.

[HD MOVIE SOURCE]

Sounds like a good conversation, thank you.