Google VP9 "Next Generation Open Video" information posted

[Jamaika]

why such poor command lines ?
I don't know better command lines. I look for better records with samples. I hope that not offended someone my opinion about the codec VP10. It doesn't interest me "pass = 2" for core i8.

[hajj_3]

new interview article about VP10 with a google manager: http://www.cnet.com/news/inside-googles-master-plan-for-faster-sharper-streaming-video/

[foxyshadis]

Unfortunately it's not a technical interview, it's mostly about HEVC's patent problems. I'd like to see an overview of what they're currently putting in it, and how much it helps.

[hajj_3]

Unfortunately it's not a technical interview, it's mostly about HEVC's patent problems. I'd like to see an overview of what they're currently putting in it, and how much it helps.

i think the most interesting part is that it won't be ready until the end of 2016. Google had previously had said that they would release a new version of VP every 18 months, vp10 would therefore have been released 6 months ago, so vp10 will be 1yr and 9 months ish later than they had previously envisioned. Daala is supposed to have an alpha version available at the end of this year. It is possible that the IETF codec by mozilla/cisco could be finished before vp10. Lets hope that google joins the open IETF codec instead as having 2 codecs would be annoying.

[LigH]

... the IETF codec by mozilla/cisco ...

Do you mean Thor?

[mandarinka]

i think the most interesting part is that it won't be ready until the end of 2016. Google had previously had said that they would release a new version of VP every 18 months, vp10 would therefore have been released 6 months ago, so vp10 will be 1yr and 9 months ish later than they had previously envisioned. Daala is supposed to have an alpha version available at the end of this year. It is possible that the IETF codec by mozilla/cisco could be finished before vp10. Lets hope that google joins the open IETF codec instead as having 2 codecs would be annoying.

I doubt netVC will be here earlier than in two years. Look how long Opus standardization took. Daala was supposed to be roughly done almost by this time, according to original plans :)

The plan to make new VP codecs ever 18 months was/is not real I think, I think they later said it was just misinterpreted/misquoted.

[Jamaika]

Google’s new VP10 algorithm to cut online video data in half… again (http://www.htxt.co.za/2015/09/01/googles-new-vp10-algorithm-to-cut-online-video-data-in-half-again/)
I don't know what algorithm is described and why it will not be under the Youtube ? What sort of quality improvement is mentioned ?
Can someone explain?
There is a catch, however; VP10 is only expected to be up and running by the end of 2016, after which it will be up to non-YouTube online video platforms to integrate it into their services.
Advertising codec VP10:
http://otavafiles.fi/plaza/s/f/editor/images/20150819vp10.jpg

[dapperdan]

Do you mean Thor?

Both Mozilla's Daala and Cisco's Thor have been contributed towards the IETF's NetVC codec effort to standardize a new royalty free codec.

Presumably, depending on how the process goes, independent development of either (or both) of those codecs would be discontinued in favour of working on the resulting codec.

[dapperdan]

And relevant to that last point, it looks like Google, Microsoft, Amazon, Netflix, Intel and others have joined in on the fun:

https://blog.mozilla.org/blog/2015/09/01/forging-an-alliance-for-royalty-free-video/

"Things are moving fast for royalty-free video codecs. A month ago, the IETF NETVC Working Group had its first meeting and two weeks ago Cisco announced Thor. Today, we’re taking the next big step in this industry-wide effort with the formation of the Alliance for Open Media. Its founding members represent some of the biggest names in online video, such as Netflix, Amazon, and YouTube, multiple browser vendors including Mozilla, Microsoft, and Google, and key technology providers like Cisco and Intel. The Alliance has come together to share technology and run the kind of patent analysis necessary to build a next-generation royalty-free video codec."

[LigH]

[img] {more or less pretty flowers}

:rolleyes: An over-simplified and idealized schematic presentation. How dubious.

[foxyshadis]

http://otavafiles.fi/plaza/s/f/editor/images/20150819vp10.jpg

Oooh, it's the difference between QCIF, SD, and 4K (conveniently skipping HD), but I'm not sure what that has to do with the VPx family.

[dapperdan]

Possibly as a result of the new royalty-free codec alliance thing, Microsoft has just commited to adding VP9 to Internet Explorer:

http://dev.modern.ie/platform/status/vp9videocodec/

Current status: "In Development"

[benwaggoner]

So, where the heck can I get a current Windows binary for vpxenc? I see it's in Staxrip and ffmpeg, but is there an actual .exe anywhere?

[LigH]

The media-autobuild_suite by jb_alvarado will build it for you. I just don't have my development PC available right now, so not before tomorrow...

[Jamaika]

So, where the heck can I get a current Windows binary for vpxenc? I see it's in Staxrip and ffmpeg, but is there an actual .exe anywhere?
Someone upset? I didn't dare ask such questions.
Selur adds codec vpxenc v1.4.0-1171. Currently codec vp10 has been turned off. There isn't need to turn it on
Possibly as a result of the new royalty-free codec alliance thing, Microsoft has just commited to adding VP9 to Internet Explorer:
How install codec vpx to Edge?

[Nintendo Maniac 64]

And relevant to that last point, it looks like Google, Microsoft, Amazon, Netflix, Intel and others have joined in on the fun:

https://blog.mozilla.org/blog/2015/09/01/forging-an-alliance-for-royalty-free-video/

Seems to me like they're not happy about the whole "HEVC Advance" thing.

[VDO2015]

With MPEG-LA announcing a call for patents for MPEG-DASH, I think that AOMedia's appetite is greater than codecs.


Sent from my iPad using Tapatalk

[Reel.Deel]

So, where the heck can I get a current Windows binary for vpxenc? I see it's in Staxrip and ffmpeg, but is there an actual .exe anywhere?

Here:VP10 code has been merged to the main branch (https://github.com/webmproject/libvpx/commits/master) few days ago...

Here are some fresh windows builds for testing : x86 (http://www.mediafire.com/download/7z88x93rtk4tkgg/vpx-1.4-1119_x86_20150815.7z)|x64 (http://www.mediafire.com/download/cdjkpvoaa7olxko/vpx-1.4-1119_x64_20150815.7z)

There's also older versions (http://forum.doom9.org/showthread.php?t=165839&page=32#post1716013) earlier in the thread.

[Koti]

vpx.enc built from /libvpx/master @ e5732bc (x64) (win64-vs11)
includes vp10

https://www.mediafire.com/?mcbtahae8dmjqx4

[LigH]

A pity the media-autobuild_suite seems to be broken today, should not have updated. Maybe I can downgrade from the repo.

VP10 sources may have got merged. But we already guessed that they are not yet completely functional. Rather a preparation of required changes for structures.

[LigH]

Apparently, the VP10 codec has to be enabled before compiling. Unfortunately I don't understand how, the "configure" script keeps resetting to the default disabling VP10 when I run the media-autobuild_suite, even if I add switches enabling VP10 explicitly.

[Kurtnoise]

vp10 is not yet ready for testing...that's why this has been disabled now whatever you run in configure.

[MoSal]

Untested win64+win32 binaries with VP10 enabled.
(Built under GNU/Linux with: mingw-w64 4.0.4, GCC 5.2.0).

https://archive.org/download/Vpx-win32win642015.09.03/vpx-win32+win64(2015.09.03).zip

[leonccyiu]

If google are going to release a new codec roughly every 18 months, then what about hardware acceleration? We don't even have vp9 yet which may not arrive until pascal or kaby lake next year at the earliest, would vp10 build upon vp9 and therefore allow a hybrid decode where the vp9 hardware decoder parts which can be used are used in conjunction with the cpu?

[hajj_3]

If google are going to release a new codec roughly every 18 months, then what about hardware acceleration? We don't even have vp9 yet which may not arrive until pascal or kaby lake next year at the earliest, would vp10 build upon vp9 and therefore allow a hybrid decode where the vp9 hardware decoder parts which can be used are used in conjunction with the cpu?

there will be no VP10 codec anymore, there will be a codec created by multiple companies instead.

[foxyshadis]

there will be no VP10 codec anymore, there will be a codec created by multiple companies instead.

That's not confirmed, Google may well simultaneously develop VP10 and the Open Media codec, or whatever they end up calling it. VP10 is already in progress, while the Open Media will be a larger fusion of different efforts and take a while. In fact, as far as I can tell Google has made no public comment on the Alliance at all.

[BadFrame]

That's not confirmed, Google may well simultaneously develop VP10 and the Open Media codec, or whatever they end up calling it.

Well, before the Alliance statement the VP10 encoder/decoder was included in the master repo checkouts (I know, as I built and tested it myself), after said statement the VP10 codec was no longer included.

Not exactly a smoking gun, but I wouldn't be surprised to find the VP10 branch shortly renamed to whatever the Alliance choose as their codec name.

In fact, as far as I can tell Google has made no public comment on the Alliance at all.

The 'Alliance' (which includes Google) had a joint statement where they outlined their goals, and that did include a statement from Google's Matt Frost:

-"Google launched the WebM Project in 2010 in the belief that web video innovation was too slow and too closed, and that broad collaboration — in the open — would fix both problems. The Alliance for Open Media is a big leap forward for these core philosophies, and we’re gratified that our AOMedia partners share this vision. Our combined strength, resources and expertise will drive the next generation of web media experiences much further and faster than WebM can do alone"

[LigH]

Because VP10 is not yet "ready", LAV Filters are not yet prepared to decode it either; just tried to encode some with MoSal's build, and MPC-HC skipped it immediately.

[GTPVHD]

http://blogs.windows.com/msedgedev/2015/09/08/announcing-vp9-support-coming-to-microsoft-edge/

[dapperdan]

I assumed Opus support would already be in place thanks to their work on WebRTC. Instead they're starting with VP9 and AAC via MSE. I guess that means either (or both) of Amazon and Netflix have said that's what they're going to serve in the short term?

Actually they just announced Dolby Digital Plus support a few months ago, basically just for Netflix, so that might be why?

[Nintendo Maniac 64]

Actually Microsoft's lack of support for Opus is quite strange since it is the evolution and the SILK and CELT codecs and SILK was developed by and for Skype.

[nevcairiel]

Microsoft is strongly considering Opus, and judging from their roadmap priority, it should probably happen eventually.
http://dev.modern.ie/platform/status/opusaudiocodec/

[BadFrame]

Because VP10 is not yet "ready", LAV Filters are not yet prepared to decode it either; just tried to encode some with MoSal's build, and MPC-HC skipped it immediately.

VP9 was not yet "ready" when they included in the master branch and there wasn't any decode support in any players.

This is why you have vpxdec, which allows you to decode the stream into a lossless format so that you can see the result.

So whatever the reason VP10 was pulled from master, I'm pretty certain that's not it (as it certainly was not the case with VP9).

[Kurtnoise]

http://blogs.windows.com/msedgedev/2015/09/08/announcing-vp9-support-coming-to-microsoft-edge/

Microsoft is strongly considering Opus, and judging from their roadmap priority, it should probably happen eventually.
http://dev.modern.ie/platform/status/opusaudiocodec/

Only available through the browser ? What a shame...

[mzso]

Only available through the browser ? What a shame...

Since that website is for Edge it's not surprising they don't mention anything else. Who knows what they're planning for windows...

[Jamaika]

Since that website is for Edge it's not surprising they don't mention anything else. Who knows what they're planning for windows...
It is now lock decoders for "Films and TV" and "WMP". The security don't omit in a simple way, eg. registering in Windows decoder MPC-BE.
I don't expect too rapid implementation VP9. So, I installed the player MPC.

[dapperdan]

I noticed this page got updated recently:

http://compression.ru/video/codec_comparison/call_for_codecs_15.html

The report (which I think was originally due last month) is now due for public release in a couple of weeks time.

[Nintendo Maniac 64]

Darn, too bad it seems that the tests were confirmed before the Broadwell and Skylake launched...

[Jamaika]

Who knows functions VP9 and X264?
Is the function --tune-content screen equivalent --tune stillimage?

[foxyshadis]

Who knows functions VP9 and X264?
Is the function --tune-content screen equivalent --tune stillimage?

It's closer to --tune animation, being designed for lots of sharp edges. But I don't know how it performs relatively.

[mandarinka]

https://www.youtube.com/watch?v=gkz1ZvejmEc

Talk about stuff intended or tested/investigated for VP10 from VDD 2015. Apparently the final format is fairly remote at this point in time, the slides also mention that some "new big ideas" are needed for it in addition to conventional boosts like 64x64 transforms, better intra, and raised internal and/or real bitdepth.

[Nintendo Maniac 64]

Here's another video from the same event that compares VP9 encoding and decoding performance to AVC and HEVC:
https://www.youtube.com/watch?v=_Q6J2_nvLSI

On a similar note, here's something interesting: the 1080p VP9 YouTube encode for the VP10 presentation mandarinka linked to is 206MB while the 1080p AVC YouTube encode is 806MB! And it's not a one-time fluke either - the Daala and Thor videos at the same event all have a similar filesize discrepancy, yet for the video I just linked above the filesize difference for 1080p is "only" 273MB (VP9) vs 457MB (AVC)...

I haven't watched this video nor the Daala, Thor, or VP10 videos just yet, so at this point in time I have no idea why this considerably smaller-than-normal difference in filesize is in fact occurring.

----------------------------------------------------------------

MS Edge officially has VP9 support now (not enabled by default):
http://www.ghacks.net/2015/09/23/enable-vp9-in-microsoft-edge-for-better-html5-streaming-support/

However, the fact that the setting is located under "Media Source Extensions" makes me curious if Edge only supports VP9 via MSE...

Could someone on Windows 10 test out the VP9/WebM video from the following website in Edge and report back? For reference the WebM file only has a VP9 stream - there's no audio stream (I checked via mkvtoolnix):
http://base-n.de/webm/VP9%20Sample.html

Direct link to the video file if you need it:
http://base-n.de/webm/out9.webm

[mandarinka]

The size discrepancies might be largely unintentional. Libvpx's rate control for VP9 is very inacurate and can drastically overshoot or undershoot. Of course quality is a whole different thing, that 806MB video is going to look tons better, unless YT is totally f...ing up (which is possible of course, their encodes were never very good).

[jmartinr]

Could someone on Windows 10 test out the VP9/WebM video from the following website in Edge and report back? For reference the WebM file only has a VP9 stream - there's no audio stream (I checked via mkvtoolnix):
http://base-n.de/webm/VP9%20Sample.html
It says "Invalid Source"

[Nintendo Maniac 64]

It says "Invalid Source"

And that's with VP9 enabled in Edge, correct?

Hopefully this is just because they don't have support for the webm container yet...

[jmartinr]

And that's with VP9 enabled in Edge, correct?.
Sorry, I totally missed that.

There's no entry "Media Source Extensions" on my Edge (20.10240.16384.0). Maybe the new version is not pushed yet?

[mandarinka]

Yeah, IIRC it was only in insider builds or something like that (you don't really expect them to push it so fast, do you. I'm not sure if there has even been an Edge update on the WU at all, since the support was announced).

[Nintendo Maniac 64]

Well the article clearly states that the option is present as of Insider build 10547 in the very first sentence...

[leonccyiu]

Here's another video from the same event that compares VP9 encoding and decoding performance to AVC and HEVC:
https://www.youtube.com/watch?v=_Q6J2_nvLSI

On a similar note, here's something interesting: the 1080p VP9 YouTube encode for the VP10 presentation mandarinka linked to is 206MB while the 1080p AVC YouTube encode is 806MB! And it's not a one-time fluke either - the Daala and Thor videos at the same event all have a similar filesize discrepancy, yet for the video I just linked above the filesize difference for 1080p is "only" 273MB (VP9) vs 457MB (AVC)...

I haven't watched this video nor the Daala, Thor, or VP10 videos just yet, so at this point in time I have no idea why this considerably smaller-than-normal difference in filesize is in fact occurring.

----------------------------------------------------------------

MS Edge officially has VP9 support now (not enabled by default):
http://www.ghacks.net/2015/09/23/enable-vp9-in-microsoft-edge-for-better-html5-streaming-support/

However, the fact that the setting is located under "Media Source Extensions" makes me curious if Edge only supports VP9 via MSE...

Could someone on Windows 10 test out the VP9/WebM video from the following website in Edge and report back? For reference the WebM file only has a VP9 stream - there's no audio stream (I checked via mkvtoolnix):
http://base-n.de/webm/VP9%20Sample.html

Direct link to the video file if you need it:
http://base-n.de/webm/out9.webm

While I get Invalid source for the link you sent me, and the direct video link gets downloaded to the hard drive, I can confirm that after enabling vp9 in Microsoft Edge on Insider Build 10547, YouTube uses vp9.

I can take a screenshot and upload it if you'd like.

Microsoft Edge is quite CPU intensive, I am using a Pentium G3258, with several tabs opens, CPU usage hovers around 50%, on avc on YouTube, CPU usage doesn't vary much and I can't hear the CPU fan much, but once I switched to VP9, the CPU fan kicked in, so I definitely know it's VP9.

[leonccyiu]

The size discrepancies might be largely unintentional. Libvpx's rate control for VP9 is very inacurate and can drastically overshoot or undershoot. Of course quality is a whole different thing, that 806MB video is going to look tons better, unless YT is totally f...ing up (which is possible of course, their encodes were never very good).

Actually, I prefer the VP9 version, I used pot player to switch between the two streams, I notice less artefacts in the shadow detail of the mans face.

If anything though, YouTube's bit-rates are far too low, 1080p looks blurred, watching 4k downscaled looks like proper 1080p. I notice that for 4k, the situation is reversed, the vp9 versions are more often larger than the h264 versions.