Besides all the fancy marketing lines we keep hearing especially around here for months now is there any objective benchmark methodology that allows us to compare AVC/h.264 with VC-1?

One problem about this surely is that we are talking "moving targets" here - the implementations and encoder applications of both codec standards are advancing and improving all the time - making it difficult to compare them without being open to the usual "but our solution has advanced alot since then..." argument.


Ben Waggoner (Microsoft Codec group) is pretty confident about VC-1 and has offered on several occasions to take real-world encoding challenges against the competition - which seems to be happening right now in the following Doom9 thread:

http://forum.doom9.org/showthread.php?p=1028887#post1028887

OPSNR and SSIM are probably the most objective and widely accepted encoding quality tests/indicators available today... "higher is better" = less deviation from the source.
(In short they are the next best thing after rdjam's highly scientifical VC-1 vs. AVC screenshot comparisons - from different movies mind you :p )

See for yourself how AVC does against VC-1 and Mpeg2 in several different bitrate scenarios in this initial review.

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

Ben's challenge:
http://forum.doom9.org/showthread.php?p=1029298#post1029298

Download both Ben's VC-1 file and the according AVC/h.264 file and compare for yourself.

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

Also discussed @ avsforum here:

http://www.avsforum.com/avs-vb/showthread.php?t=882846


For everybody who is not aware of Doom9 - this is THE place on the net for any compression related discussions. To call most Doom9 members encoding "enthusiasts" is a gross understatement = they know very well what they are talking about.

First results may come as a big surprise for anybody whose primary source of codec/video compression related information and knowledge has been the avsforum (the insider aka marketing thread in particular).


I sincerely hope we can finally bring SCIENCE back to the avsforum.

I don't think it's worth it. I have encoded with h264 and WMV9, which is basis for VC-1. The bottom line is the picture quality is potentially the same for either one. The difference is the encoding speed, i.e. one pass, two pass, and which one does it at the lower bit rate.

Yes, I am aware of doom9 and videohelp.com in another pretty good one.

This is bragging rights, nothing more.

Thanks for the heads up on what's been happening with the codec challenge; it'll be interesting to see what arguments are made going forward as some of the insinuations made prior come under increasing challenge.

I don't think it's worth it. I have encoded with h264 and WMV9, which is basis for VC-1.
Just a minor point of clarification - WMV9 is a VC-1 implementation, not the basis of it. Any WMV9 or WMV9 AP stream is VC-1 compatible, and any VC-1 stream is comptible with the WMV9 or WMV9 AP decoder (assuming you get it into a compatible file wrapper).

X264 is an awesome AVC encoder. I've had great success encoding 24P HDV material from a Canon HV20.

Now, how does X264 compare to the Sony/Panasonic encoders and can it be used to author Blu-Ray/HD-DVD discs?

Ben:

Will you be encoding your own stream with your "latest and greatest" using Sagittaire's profiles and posting the results?

Anything like this that gets the attention of MS or AVC companies and pushes them to improve the quality of their encodes is good news for us HD disk enthusiasts.

If you have a good raw stream with film grain in it, why don't you make it available, and propose an encoding challenge for that stream to Sagittaire as well, and see how you stack up to H264?

This challenge will end up being just as useless as ANY other challenge in the past that had Sagittaire as one of the contributors.

Reason:

he constantly keeps changing the rules to these challenges whenever the results are not as favorable to his position as he anticipated.

Just witness that he did that again today.

(In short they are the next best thing after rdjam's highly scientifical VC-1 vs. AVC screenshot comparisons - from different movies mind you )Heh - fame and fortune seem to follow my "scientifical" ventures :)

Tell you what tho - there are so many comparison threads on the forum threads that, whether same or different films, one can see which codecs have produced the most stellar products. Right now, that's King Kong, IMO.

The sharpness and levels of detail shown in KK in the DU thread by FrancescoP have outstripped almost everything else. Forget 720p for a minute - In my 1440x1080 DU simulation, the difference in detail was still apparent.

Heh - fame and fortune seem to follow my "scientifical" ventures :)

Tell you what tho - there are so many comparison threads on the forum threads that, whether same or different films, one can see which codecs have produced the most stellar products. Right now, that's King Kong, IMO.

The sharpness and levels of detail shown in KK in the DU thread by FrancescoP have outstripped almost everything else. Forget 720p for a minute - In my 1440x1080 DU simulation, the difference in detail was still apparent.

rdjam,

we certainly agree about many stellar VC-1 encodes and King Kong in particular - it is still one of my preferred demo transfers and easily one of the best (read: most detailed, certainly not most artifact free...) HD available.

But you will certainly agree that for King Kong one of the very best sources available in the industry today was used.

A given codec/encoder can never MAKE a given movie, all it "can do" is BREAK it by introducing artifacts and/or removing detail. In short "King Kong" would look outstanding no matter if (high bitrate, >25MBit/s avg.) Mpeg2 or AVC or VC-1 is used.

Last time I checked the "s" in AVSforum stands for "Science". So let's put all the marketing agenda's and shallow "Movie A using codec B looks better than movie C using codec D and alot better than movie E using the implementation F of codec G" arguments behind us (at least for this single thread...) and let's try to look at some "scientifical evidence" for once.

That's why we have OPSNR and SSIM2 benchmarks.

This challenge will end up being just as useless as ANY other challenge in the past that had Sagittaire as one of the contributors.


Well it's not useless simply because MicroSoft use for example metric for ratio comparison between MPEG2, VC1 and H264:
http://www.microsoft.com/windows/windowsmedia/howto/articles/vc1techoverview.aspx

VC-1 achieves clearly superior quality to MPEG-2 at comparable bit rates, and has been judged superior to H.264 in several independent studies.

Measuring the quality of a video codec is not easy, because the reconstructed image is not meant to be identical to the original. Ideally, only information that is perceptually irrelevant will be lost in the compression/decompression process, but what counts as "irrelevant" depends on the viewer's subjective response.

One useful objective metric is the peak signal-to-noise ratio (PSNR) plotted against bit rate. PSNR is the ratio between the maximum value of a signal (255 for 8-bit video) and the quantization noise. A higher PSNR indicates a less noisy signal. For any codec, PSNR is expected to increase at higher bit rates, because higher bit rates translate to less aggressive compression. Thus, a graph that plots PSNR against bit rate shows the performance of the codec over a range of compression settings.

In Microsoft's own internal tests, VC-1 performs 2 to 3 times better than MPEG-2. In other words, to achieve a given PSNR, MPEG-2 requires a bit rate up to 3 times higher than VC-1. These results were measured using both low-motion and high-motion video sequences. Microsoft also compared VC-1 with H.264 and found that both codecs have comparable performance when PSNR is plotted against bit rate.



he constantly keeps changing the rules to these challenges whenever the results are not as favorable to his position as he anticipated.

Well I'am really curious ... you have a particular example ... ???


Just a minor point of clarification - WMV9 is a VC-1 implementation, not the basis of it. Any WMV9 or WMV9 AP stream is VC-1 compatible, and any VC-1 stream is comptible with the WMV9 or WMV9 AP decoder (assuming you get it into a compatible file wrapper).

It's true but like you say there are major difference between WMV9 AP from WM11 SDK and VC1 from PEP. Moreover it's in practice impossible to make HDDVD/BD compliant stream with WMV9 AP implementation.


we certainly agree about many stellar VC-1 encodes and King Kong in particular - it is still one of my preferred demo transfers and easily one of the best (read: most detailed, certainly not most artifact free...) HD available.

KK master is perhaps a very clean source easy to encode. If you don't have the equivalent AVC and MPEG2 encoding you can't conclude.

I remember reading those threads a long while back and realising it was going to be very difficult to reach any consensus as there were so many unknowns.

I'll have to read up again but from earlier reading at Doom9 I'd have to agree with kschmit2. Unless things have changed of course.

Also, while I agree that Doom9 is "THE" place for compression, that does not mean it is all completely unbiased info there. There are a few level headed people whereby if you ask them what codec is best they won't tell you a straight answer because they know full well that the implementation and skill of the compressionist matter a great deal. One thing you can generalise is that there are a great deal of x264 fans at that site, and that they don't get as much info as we do from insiders here, hence a very strong hatred of MS (or M$ as many of them refer to it) exists.

I've even seen cases where at certain bitrates XviD looks better than x264 and you could see people scrambling to find excuses. There are cases where x264 measured better but screencaps clearly showed otherwise. OPSNR and SSIM are great, but don't tell the whole story. Its unfortunate that we have no real way of measuring this stuff.

It's true but like you say there are major difference between WMV9 AP from WM11 SDK and VC1 from PEP. Moreover it's in practice impossible to make HDDVD/BD compliant stream with WMV9 AP implementation.

Yes, they're different implementations, but they're still both compliant VC-1 implementations. Our codec efforts will be converging, so a single library can be used to create both WMV, HD DVD, BD, and IPTV streams. Still, there will be different optimizations for each target environment.

One thing you can generalise is that there are a great deal of x264 fans at that site, and that they don't get as much info as we do from insiders here, hence a very strong hatred of MS (or M$ as many of them refer to it) exists.

Its unfortunate that we have no real way of measuring this stuff.

There are fans of "open codecs" that can be used across different operating systems. And if you go back in Doom9´s history, Microsoft have had a history of producing the bad video compression. They never landed on top in any of the classic "DiVX" duels. VC-1 is the fruits of many years work and boy did it take Microsoft a long time :-)

As for your last sentence, your own eyes are the best judge.

As for your last sentence, your own eyes are the best judge.

I agree... but that's not what this thread is about.

Zodiaque, why not post under your regular handle (Sagittaire)?

Key problem here is the source - it's too clean. We need to find some short sample of something on film.

As for eyes-vs-metrics, downloading the finished streams might provide some interesting discussion points. Metrics don't always rate visible artifacts quite as heavily, since it's a specific thing we can highlight - the screenshot threads use this approach. On the other hand, metrics are pretty good at summarising how much overall detail is consistently retained, which can have more of a subconscious effect imo (ie mainting the directors' artistic intent).

Personally I find SSIM outstanding in how it correlates with visual quality. It maybe needs more work on individual frame quality drops, but as an average it really does do the job well.

I agree... but that's not what this thread is about.

Stream will be available for all bitrate. Download and compare with your eyes is possible.


Zodiaque, why not post under your regular handle (Sagittaire)?

I don't remember any more ... perhaps simply because Sagittaire name is not free on this forum.


Personally I find SSIM outstanding in how it correlates with visual quality. It maybe needs more work on individual frame quality drops, but as an average it really does do the job well.

OPSNR is really usefull for find Rate Control Problem. In fact for good result you must use OPSNR, SSIM and graph PSNR and confirm with visual test.

Also, while I agree that Doom9 is "THE" place for compression....... There are a few level headed people whereby if you ask them what codec is best they won't tell you a straight answer because they know full well that the implementation and skill of the compressionist matter a great deal.
Well then, would you agree that the compressionists that take up the challenge on Doom9 are much more likely to be of the same expertise as the typical studio inhouse compressionist or post house compressionist that provide real world encodes which the public sees on a daily basis rather than someone who is the World's Greatest Compressionist ?

(scroll down to Comments and Things Worth Noting)
http://forums.creativecow.net/showprofile/117/3657

Who of course is using the world's greatest codec to begin with.

Key problem here is the source - it's too clean. We need to find some short sample of something on film.


An alternative to this, that can be more "controlled" and thus quantifiable, could be to use that same source that is "too clean", but add a controlled amount of noise to it (some % rating), and then do the encoding evaluation. That way you can explore a range of source conditions from absolutely clean to known, varied degrees of noisy.

Well then, would you agree that the compressionists that take up the challenge on Doom9 are much more likely to be of the same expertise as the typical studio inhouse compressionist or post house compressionist that provide real world encodes which the public sees on a daily basis rather than someone who is the World's Greatest Compressionist ?

(scroll down to Comments and Things Worth Noting)
http://forums.creativecow.net/showprofile/117/3657

Who of course is using the world's greatest codec to begin with.

Don't know what you are trying to achieve with this post. I would say that there are many people on Doom9 who are able to create encodes at the same level of expertise as those who work in studios etc. What I mean by this is that there will be great compressionists who are hobbyists and there will be not so great compressionists who do it for a living.

As for the stab at Ben, I'm sure he's used to it. Read the forums and you'll know what I mean. If you view it with as little bias as possible you will see plenty of people speak passionately against MS/VC1 (even though their arguments are based on their own assumptions) and you won't find Ben being miffed about it, responding to their posts and not resorting to attacks. Given this candid nature I'm willing to bet that the self appointed title of "World's Greatest Compressionist" is all in good fun. Kinda like: http://www.thebestpageintheuniverse.net/

As for the world's greatest codec, do read the doom9 forums as Ben has never stated as such. When people "crap" all over VC-1 encodes, you won't hear a "you're stupid and don't know how to use it" comment, instead constructive posts are made to ascertain settings, hardware setup etc.

My main point of concern is that one of the moderators clearly has a preference for x264 and clearly does not like MS as a company. Participation from someone who is a moderator does not strip them of their own opinions, but it does mean they have a certain responsibility to steer discussion away from potential flame wars. In an objective comparison of VC1 vs x264 what does it matter what you think of the company? That's like saying PS3 sucks compared to Xbox360 because Sony used rootkits once. It really makes you appreciate the level of impartiality displayed by the AVS moderators.

Cheers...
Duy-Khang Hoang

Duy-Khang,
I didn’t know the question was that difficult to interpret.
Allow me to rephrase.
See below.

Also, while I agree that Doom9 is "THE" place for compression..........There are a few level headed people whereby if you ask them what codec is best they won't tell you a straight answer because they know full well that the implementation and skill of the compressionist matter a great deal.
Well then, would you agree that the compressionists that take up the challenge on Doom9 are much more likely to be of the same expertise as the typical studio inhouse compressionist or post house compressionist that provide real world encodes which the public sees on a daily basis rather than someone who has the credentials of Ben W.--completely disregarding the self-anointed title, whether it be assigned in jest or with a hint of seriousness ?

Remember, apparently the only thing that is physically sent up to Redmond for Ben (or his subordinates) to do is difficult material that WB inhouse compressionists or Uni post house compressionists are having a helluva a time to deal with. As I believe, he’s claimed somewhere on this forum that Microsoft doesn’t do end-to-end encodes on any titles……..at least not anymore, I’m unsure of the temporal aspect of his previous statement.

I’m really just trying to get a handle on how important as you say…… the “skill of the compressionist” will be in this particular challenge on Doom9.
It’s not a “stab” at Ben at all. I’m just wondering if the other guys even have a chance.

And as far as moderators go, inherent bias is not restricted to any one forum on the internet but, it does seem like things are improving recently. :)

As for the world's greatest codec, do read the doom9 forums as Ben has never stated as such.
I never said he did, but do read the AVS forum where it has been implied repeated ad nausea for months, years? by his colleague.

(scroll down to Comments and Things Worth Noting)
http://forums.creativecow.net/showprofile/117/3657
That cracked me up! I must have stuck that in there years ago and forgot about it. It's a reference to a .sig I used for about six weeks in 2001 that folks in the scene still josh me about.

I see that my title is about three reorgs old as well :).

Who of course is using the world's greatest codec to begin with.
That's too broad a claim for me to make. I will say that I think that VC-1 is the best codec for HD encoding of film source content available today.

There's other environments and scenarios where other codecs can be better, and I can certainly imagine ways a "VC-2" could be better yet. But I'm happy with what VC-1 is for the uses relevant to this forum, and where we're going with our implementation of it.

So if VC1 is proven to be inferior to AVC then what's the big deal? It seems that there is an argument being made repeatedly to embrace BD and burn HD DVD because in some tests HD DVD doesn't perform as well as AVC.

Have we seen a 24fps film source used as a test? Anyone who knows even a bit about encoding should know that animation or video will not encode the same as 24 fps film.

It just seems that we are getting another pitch to abandon HD DVD and embrace BD since some BD studios are releasing with AVC now.

So if VC1 is proven to be inferior to AVC then what's the big deal? It seems that there is an argument being made repeatedly to embrace BD and burn HD DVD because in some tests HD DVD doesn't perform as well as AVC.

Have we seen a 24fps film source used as a test? Anyone who knows even a bit about encoding should know that animation or video will not encode the same as 24 fps film.

It just seems that we are getting another pitch to abandon HD DVD and embrace BD since some BD studios are releasing with AVC now.
Boo hoo.

I don't want to play too much into your format war hands j, but the way I see it, some insiders have bashed AVC in lieu of their favored VC-1 with impunity. This is sour grapes on your part. VC-1 was your big, Holy Grail and now it's gone. I recommend acceptance vs. denial. This thread just answers some questions that were raised by HD DVD insiders and fans. Just because you don't like the answers...

There's other environments and scenarios where other codecs can be better, and I can certainly imagine ways a "VC-2" could be better yet.
There's already a VC-2 and VC-3.

http://www.ebu.ch/CMSimages/en/EBU-2006ProdTechnoSeminar-Report_FINAL_tcm6-43103.pdf

Ron

Ron,
I wish you would post more often and spend less time at the "Gentlemen's Club." :)

That cracked me up! I must have stuck that in there years ago and forgot about it. It's a reference to a .sig I used for about six weeks in 2001 that folks in the scene still josh me about.

Actually Ben if I’m not mistaken, that handle began earlier than 2001 ?……….
http://listproc.ucdavis.edu/archives/digvid-l/digvid-l.log9901/0010.html

and has kinda stuck even on the background submitted as a presenter at various conferences over the years.

http://www.hesca.org/past_meetings/dallas/daythree.htm

https://www.cmpevents.com/DVx3/a.asp?option=G&V=3&id=228626

http://www.rcconsulting.com/Sample_Sites/stanfordmedia/sp_instructors.html

Better watch out, there’s always a faster gun in the West that comes along. ;)

That's too broad a claim for me to make. I will say that I think that VC-1 is the best codec for HD encoding of film source content available today.


Hi Ben,

I don't have the same think, (two year ago VC1 did a little better result of old H264 implementations) today,
I'm sure ( my opinion with my experience of HD movie wth grain or not) H264 is the best codec for HD encoding
and better than VC1 (I think everytime, but may be there is some specific movie with specific H264 implementation
where VC1 will do a little better result).

Like Zodiaque ( Sagittaire on Doom 9, ?), I prefer the metric and a public test, all parameters and video encoded
are available to public. Today I can say with result of first round test, H264 is really better than VC1
and the best MPEG2 implementation can make a "good fight" with VC1.


I really hope that MS team can show a different result with VC1 (because it use on all HD DVD Title, 90%).
And I hope too that some developper of CC-HDe (from CTC company) can participe at this test
( CC-HDe is the H264 encoder for Disney title, Pixar, ....).

VC-1 was your big, Holy Grail and now it's gone.

Urrrm, both formats can use AVC and VC1 equally. In fact, since AVC is probably in a stronger position relative to VC1 at lower bitrates, it's more likely to be of benefit to HDDVD than Bluray.

Please can we not make this into another tedious disc-fanboy thread? It's simply irrelevant to the point of the comparison.

Boo hoo.

I don't want to play too much into your format war hands j, but the way I see it, some insiders have bashed AVC in lieu of their favored VC-1 with impunity. This is sour grapes on your part. VC-1 was your big, Holy Grail and now it's gone. I recommend acceptance vs. denial. This thread just answers some questions that were raised by HD DVD insiders and fans. Just because you don't like the answers...

You are apparently mistaking me with someone else. I have no illusions that VC1 is the "best" codec in all situations and have stated so repeatedly in the screenshot comparison thread.

Now if you want to argue if the PQ differences between a good VC1 encode and a good AVC encode are substantial in watching a movie, then you've got a fight on your hands, because in reality they are very very close.

You are apparently mistaking me with someone else. I have no illusions that VC1 is the "best" codec in all situations and have stated so repeatedly in the screenshot comparison thread.
That's great, j. Glad to hear it. Call it a "...cried wolf" thing. Let's be honest, you're reputation is not exactly one of non-fanism, for lack of a better word and your post sure rang of good ol' jmpage to me, but... my mistake.

Now if you want to argue if the PQ differences between a good VC1 encode and a good AVC encode are substantial in watching a movie, then you've got a fight on your hands, because in reality they are very very close.
Definitely do not. I always believed that. Glad you do as well. It was a bit maddening to read some insiders' (really just one) views on the subject. Hard to imagine how he could've gotten a reputation for spouting B.S... again with impunity on a thread where he was not allowed challenge by non-insiders. I'm very glad to see "science" taking the forefront here.

Please, carry on.

That's great, j. Glad to hear it. Call it a "...cried wolf" thing. Let's be honest, you're reputation is not exactly one of non-fanism, for lack of a better word and your post sure rang of good ol' jmpage to me, but... my mistake.


Definitely do not. I always believed that. Glad you do as well. It was a bit maddening to read some insiders' (really just one) views on the subject. Hard to imagine how he could've gotten a reputation for spouting B.S... again with impunity on a thread where he was not allowed challenge by non-insiders. I'm very glad to see "science" taking the forefront here.

Please, carry on.

Any "fanboism" that I have a reputation for comes from my own personal preference for HD DVD at this time and failing to bow to the odious behavior of BD trolls far and wide.

You are apparently mistaking me with someone else. I have no illusions that VC1 is the "best" codec in all situations and have stated so repeatedly in the screenshot comparison thread.

May be I understand what you want to say: VC1 video file from HDDVD VC1 toolkit of Microsoft is nice because this toolkit have
good preprocessing software (see doc file with toolkit) to optimize the video to be encoded in VC1.

But with the same preprocess ( 4:4:4 o 4:2:2 -> 4:2:0, reduce/remove colour banding,...) H264 will be better (to my mind, we need to test).
I think the video preprocessing on CinemaCraft-HDe can do more than software HDDVD VC1 Toolkit (we need to test but 80000$).
The challenge compare the codec's efficient without video preprocessing.

You can give a video movie ("free" without problem of copyright...) to be included in codec challenge, to compare the screenshot after.



Regards,



Golgot13

There's already a VC-2 and VC-3.

http://www.ebu.ch/CMSimages/en/EBU-2006ProdTechnoSeminar-Report_FINAL_tcm6-43103.pdf

Ach, if I knew you were reading I would have picked a better example :).

VC-1++?

I just mean a theoretial future codec using VC-1 as a starting point but not backwards compatible to it.

I don't have the same think, (two year ago VC1 did a little better result of old H264 implementations) today, I'm sure ( my opinion with my experience of HD movie wth grain or not) H264 is the best codec for HD encoding and better than VC1 (I think everytime, but may be there is some specific movie with specific H264 implementation where VC1 will do a little better result).
We have different opinions. However, we have the rare pleasure here on AVS of asking emperical questions for which we can emperically test :).

Like Zodiaque ( Sagittaire on Doom 9, ?), I prefer the metric and a public test, all parameters and video encoded are available to public. Today I can say with result of first round test, H264 is really better than VC1 and the best MPEG2 implementation can make a "good fight" with VC1.
The first round VC-1 test wasn't valid for what AVS wants to know. They were done with an older implementation of VC-1 than what the studios are currently using, and with some non-optimal (and somewhat inexplicable) settings. While it's an interesting data point, it underestimates and does not reflect the tools and workflow currently in use for titles in production today.

While it's an interesting data point, it underestimates and does not reflect the tools and workflow currently in use for titles in production today.

Yes and not, because we will have more H264 solution before the end of 2007.
Today I can give some nice solutions:
- CC-HDe PC turnkey for HD DVD and BD from CinemaCraft company
- CineVision from Sonic
- BAE-VA700 H.264/AVC Encoder from Sony (can do HD DVD and BD encoding)

There are too, but only encoder (software):
- Elecard Converter Studio from Elecard
- H.264 Encoder from Mainconcept
- H.264 Encoder PC Suite from Ateme (compliant with specific hand made profil)
- Toshiba H264 encoder in CLI (but old H264 implementation = less quality)
- Panasonic H264 encoder

And some hardware development:
- VP3 (FPGA like Fathom of Inlet) from Vitec Multimedia (can make compliant HD DVD stream).


For VC1, I know PEP (I think the best solution, better than Inlet encoder), Cinevision PSE (First development of VC1 SDK of Microsoft)
from Sonic, CineVision from Sonic and Inlet solution( compliant ? I'm not sure).

A little ask, Ben:
why Sonic was the first (4 month before others companies) which received a VC1 SDK or help from Microsoft to make CineVision PSE and
why MS said at all authoring studio to choose this solution (MS have others partners which propose HD DVD solution too) ?

Any "fanboism" that I have a reputation for comes from my own personal preference for HD DVD at this time and failing to bow to the odious behavior of BD trolls far and wide.

If you talked about me, I recommand you to see my old post (first on AVS):
I was one of first supporter of HD DVD, but after some help (exactly: nothing, only wind) from some company
I change my opinion (after lot of H264 development from many companies) about VC1 and HD DVD.

H264, to my mind, it's really a codec for HD DVD. And VC1, because it optimize for high bitrate, it's a codec adapted for BluRay (it is not a joke).



Regards,



Golgot13

Yes and not, because we will have more H264 solution before the end of 2007.
Sure. The hard work for all of us codec vendors and the good news for consumers is there's fierce competition in the codec market, so we're seeing rapid innovation in tools.

A little ask, Ben:
why Sonic was the first (4 month before others companies) which received a VC1 SDK or help from Microsoft to make CineVision PSE and
why MS said at all authoring studio to choose this solution (MS have others partners which propose HD DVD solution too) ?
Sonic was able to take PEP as it is today and sell and support it. Sonic is uniquely positioned in the professional conten creation market, so there we an obvious early parter for us. The intial versions of CineVision PSE will be very much updated versions of PEP.

We have other projects ongoing to support other, broader markets.

If you talked about me, I recommand you to see my old post (first on AVS):
I was one of first supporter of HD DVD, but after some help (exactly: nothing, only wind) from some company
I change my opinion (after lot of H264 development from many companies) about VC1 and HD DVD.

H264, to my mind, it's really a codec for HD DVD. And VC1, because it optimize for high bitrate, it's a codec adapted for BluRay (it is not a joke).



Regards,



Golgot13

The comment was not directed at you Golgot, but at some other individuals here at AVS.

Sonic was able to take PEP as it is today and sell and support it. Sonic is uniquely positioned in the professional conten creation market, so there we an obvious early parter for us. The intial versions of CineVision PSE will be very much updated versions of PEP.


Yes, I know but why Sonic in first, not Inlet (first WMVHD card encoder) or Mainconcept (first VC1 implementation full compliant HD DVD or BD),
or MemoryTech (first professional HD DVD solution which work, after Toshiba solution but it's impossible to buy it), or Ulead (first HD SDK for authoring and encoding) ....



We have other projects ongoing to support other, broader markets.

Yes, I know. VC1 encoder price will down with this new solutions (it is better for professionnal market and HD DVD title will be cheap after this).
But I understand you give at others companies only a VC1 SDK, but you gave at Sonic for CineVision PSE all code (and more) of PEP encoder...


Regards


Golgot13

but you gave at Sonic for CineVision PSE all code (and more) of PEP encoder...

That is not true, they don't have the source code to PEP, just the binaries. They are simply re-selling PEP.

Any "fanboism" that I have a reputation for comes from my own personal preference for HD DVD at this time and failing to bow to the odious behavior of BD trolls far and wide.
Accepted.

Since you used quotes, I must remind you, for the record that I carefully chose the word "...fan-ism." No "...boi..." involved and no insult intended.

I'm way over this, but if you're not, please shoot me a PM if you wish to discuss this further.

Movin' on

Yes, I know but why Sonic in first, not Inlet (first WMVHD card encoder) or Mainconcept (first VC1 implementation full compliant HD DVD or BD),
or MemoryTech (first professional HD DVD solution which work, after Toshiba solution but it's impossible to buy it), or Ulead (first HD SDK for authoring and encoding)...
We work with a variety of partners in different markets, who announce stuff based on their own development timelines.

They were done with an older implementation of VC-1 than what the studios are currently using

Old implementation ... ???
VC-1 Sequential Encoder v 1.06 (7/12/2006 build). Just 8 Months. The large majority of HDDVD available today use certainely an older or this build, isn't it?


and with some non-optimal (and somewhat inexplicable) settings.

No optimal setting? All the ME/RDO search at maximum quality (complexity at maxi, chromasearch at maxi, SADT motionmatch) with default setting for DQuant and optimized setting for SSIM boost. Come on ben it's the setting that you use yourself for this source with WMV9 AP implementation. I am sure that studio (even with massive parallel encoding) don't use these setting because it's too slow: if my memory is good more than 30 000 sec (last pass) for the encoding with 15690 frames with c2d at 3.0 Ghz ... !!? Anyway if the latest VC1 implementation can make really better job ... why not take part in the challenge and make yourself the encoding. MS claim that VC1 can obtain metric really close to H264. I want see that ...


While it's an interesting data point, it underestimates and does not reflect the tools and workflow currently in use for titles in production today.

Well real studio make real encoding (HDDVD and BD) with this real source:
http://www.hddvd.de/elephantsdream/index.html

Old implementation ... ???
VC-1 Sequential Encoder v 1.06 (7/12/2006 build). Just 8 Months. The large majority of HDDVD available today use certainely an older or this build, isn't it?
Discs have already been completed with a later version.

No optimal setting? All the ME/RDO search at maximum quality (complexity at maxi, chromasearch at maxi, SADT motionmatch) with default setting for DQuant and optimized setting for SSIM boost. Come on ben it's the setting that you use yourself for this source with WMV9 AP implementation. I am sure that studio (even with massive parallel encoding) don't use these setting because it's too slow: if my memory is good more than 30 000 sec (last pass) for the encoding with 15690 frames with c2d at 3.0 Ghz ... !!? Anyway if the latest VC1 implementation can make really better job ... why not take part in the challenge and make yourself the encoding. MS claim that VC1 can obtain metric really close to H264. I want see that...
Actually, max complexity is actually not the full optimal setting, since it locks in a few parameters at less than the full value (counterintuitive, I know).

Encoding for WMV9 AP isn't the same as with HD DVD, also. Some of the parameters have different subilties than the released version.

We're heads down on a realease right now, but I'll try to get a current optimum clip availalbe as soon as possible (this came up with inconveniant timing :)).

Well real studio make real encoding (HDDVD and BD) with this real source:
http://www.hddvd.de/elephantsdream/index.html
It's an interesting test (I use it myself a bunch, certainly) but since it's CGI and full 16:9 , it's not representative of most of the source going onto disc today.

That is not true, they don't have the source code to PEP, just the binaries. They are simply re-selling PEP.


See the post of Ben which said :

The intial versions of CineVision PSE will be very much updated versions of PEP.


So, MS give code at Sonic :D
I don't think that MS make code and Sonic sell it (but may be because CineVision is from Mainconcept SDK
Sonic make a nice GUI and some optimization; and HD authoring software from Sonic don't work well, but there is lot of update ;) ).

To my mind, the best company for VC1, after MS, is Inlet (so why Inlet don't sell PEP 2 ? Why only Sonic sell it?).
But there is not lot of update for Scenarist BD (and it's hard to use it, BluPrint is the better today).





Golgot13

So, MS give code at Sonic :D
sspears and I both work on that project :)!

Sonic doesn't have the source code.

To my mind, the best company for VC1, after MS, is Inlet (so why Inlet don't sell PEP 2 ? Why only Sonic sell it?).
Inlet is a great partner for us and for VC-1. But they're focusing on different areas than Sonic (particularly live encoding, IPTV, and WMV applications in general).

Discs have already been completed with a later version.

well how many HDDVD/BD use an older or the same build. 95% (?) of the available HDDVD/BD movie ... perhaps and certainely more I think.


Actually, max complexity is actually not the full optimal setting, since it locks in a few parameters at less than the full value (counterintuitive, I know).

Difficult to use no available setting. Anyway complexity "5" setting will save perhaps something like 1% or 2% of size ... perhaps 3%. There are certainely a good reason if it's not actived by the MS dev (certainely a time overkill with very little quality improuvement like I say).


Encoding for WMV9 AP isn't the same as with HD DVD, also. Some of the parameters have different subilties than the released version.

Certainely false for ME/RDO setting. Better complexity search produce always better "quality" (for metric here). Certainely true for DQuant and other HVS setting. For this profil we test all the possible combinaison for each parameters for produce the best possible SSIM.


We're heads down on a realease right now, but I'll try to get a current optimum clip availalbe as soon as possible (this came up with inconveniant timing :)).

:)

sspears and I both work on that project :)!

Sonic doesn't have the source code.


The price of CineVision PSE is only for Sonic support (no longer MS support you said).
Sonic is very happy (you make code, they sale it).

And why some other company can not sell it?
Eg: Ulead or MemoryTech or NetBlender (I understand Ulead and NetBlender make software for BD and HD DVD
like Sonic but MemoryTech make only software for HD DVD).....

First HD DVD authoring companies in Europe (some in US, because Toshiba authoring software) use (or use only ?)
MemoryTech authoring software. Sonic is your partner but MemoryTech not ?

I see and test CineVision PSE, the video quality is same (very few optimization quality, tests at 10Mbps and 18Mbps)
than from PEP encoder.

The price of CineVision PSE is only for Sonic support (no longer MS support you said).

MS still directly supports several of the compression houses. Sonic is only supporting the newer ones. By newer, I mean those that Microsoft was not working with before Sonic started re-selling PEP.

well how many HDDVD/BD use an older or the same build.

No one uses vc1_enc.exe to encode movies. This was my test app that was used in the early days of codec development, before PEP existed. It only provides a subset of the PEP encoder settings. I don't know why it is still included and probably won't be with the next release.

There is no complexity 5 in PEP. 5 and 4 are the same, except 5 turns on a bunch of tools by default. All of those are directly exposed to the user in PEP. Version 1.06 is the last official version released.

1.07 is now being tested by a few facilities and will go to a much wider audience in the near future. The first title that used 1.07 will be released this month.

MS still directly supports several of the compression houses. Sonic is only supporting the newer ones. By newer, I mean those that Microsoft was not working with before Sonic started re-selling PEP.



No one uses vc1_enc.exe to encode movies. This was my test app that was used in the early days of codec development, before PEP existed. It only provides a subset of the PEP encoder settings. I don't know why it is still included and probably won't be with the next release.

There is no complexity 5 in PEP. 5 and 4 are the same, except 5 turns on a bunch of tools by default. All of those are directly exposed to the user in PEP. Version 1.06 is the last official version released.

1.07 is now being tested by a few facilities and will go to a much wider audience in the near future. The first title that used 1.07 will be released this month.


Wow. Anyone else feel left out? :p :D

MS still directly supports several of the compression houses. Sonic is only supporting the newer ones. By newer, I mean those that Microsoft was not working with before Sonic started re-selling PEP.


Strange, because lot of authoring companies in europe received a email at end of February (or begining of March) about PEP.
This email said that PEP was stop, there was no support and it was replace by Cinevision PSE (which it is not free).
The company, where I work, had no support after March (before NAB) and we never receive a new version of PEP (after the HD DVD VC1 Toolkit CD)...
May be, MS support only first compression houses (which received many months advanced.... :mad: ) and CineVision PSE is free for this companies
(I will not surprise it is true !!! European commission will be very interested by this !!!! )

The only facility I have worked with in Europe, and continue to work with, is CMC/DMP. Even though they purchased from Sonic, I will still be supporting them. Same goes for MemoryTech in Japan, Deluxe, Technicolor, GDMX and 1k on the US.

Would you say that because the eval version expired, and you are expected to purchase a full version, that you are going to use other codecs for projects? (assuming you used VC1 before it expired.)

Last time I checked the "s" in AVSforum stands for "Science". So let's put all the marketing agenda's and shallow "Movie A using codec B looks better than movie C using codec D and alot better than movie E using the implementation F of codec G" arguments behind us (at least for this single thread...) and let's try to look at some "scientifical evidence" for once.



Good luck with that. There are about as much impartial scientific researchers in this area on this forum that don't have an agenda or favorite, as there are for global warming.

The first title that used 1.07 will be released this month.Can you give us a hint? :)

Good luck with that. There are about as much impartial scientific researchers in this area on this forum that don't have an agenda or favorite, as there are for global warming.
You can go anywhere there are competing technologies in any field, where the other party only knows what has been revealed to them, and you will get the same technical bickering. Guaranteed.

No one uses vc1_enc.exe to encode movies. This was my test app that was used in the early days of codec development, before PEP existed. It only provides a subset of the PEP encoder settings. I don't know why it is still included and probably won't be with the next release.

vc1_enc.exe use 1.06 core too and it's exactly the same date of compilation for pep.exe, pepworker.exe, pepwsvc.exe or vc1_enc.exe. vc1_enc.exe seem produce the same result frame by frame than pep.exe. But it's possible to compare pep.exe 1.06 result and vc1_enc.exe 1.06 result with the same setting if you want. No one uses vc1_enc.exe but vc1_enc.exe don't have eval expiration like pep.exe. Moreover CLI profil for vc1_enc.exe is by far better for test than heavy xml project for pep.exe. Test for compare pep.exe 1.06 and vc1_enc.exe 1.06 in progress ...


There is no complexity 5 in PEP. 5 and 4 are the same, except 5 turns on a bunch of tools by default. All of those are directly exposed to the user in PEP. Version 1.06 is the last official version released.

I know that complexity at 4 is the maximum complexity level available with pep. Benwaggoner speak about a possible higher complexity level and not me. Perhaps that this level is available for WMV9 AP (6 quality level [0-20-40-60-80-100] ) and not for pep (5 quality level [0-1-2-3-4] ).

Anyway if the latest VC1 implementation or pep.exe can make really better job ... why not take part in the challenge and make yourself the encoding. MS claim that VC1 can obtain metric really close to H264. I want see that... :)

(6 quality level [0-20-40-60-80-100] ) and not for pep (5 quality level [0-1-2-3-4] ).

Since starting on PEP two years ago, I have not followed the changes made to the DMO used by the GUI encoder. At one point 80 and 100 were the same complexity level. I am not sure if it is still that way. Complexity 5 was introduced to enable certain features so that regkeys did not have to be used. Since the GUI encoder app has not been worked on in over 3 years, regkeys are the only way to enable new features when you swap out the DMO.

If anyone is involved with the challenge, it would probably be Ben or Alex as I busy on the next codec release for PEP. As well as encoding a few of projects myself.

for vc1_enc.exe is by far better for test than heavy xml project for pep.exe.

I hear ya. I used to be able to write batch files to encode 1000s of combinations and now I have to create xml files and I just can't do as much in the same time frame. vc1_enc.exe also never provided 3rd pass. (what we call re-encoding)

Can you give us a hint?

8/28 and I wanted to ensure it looked the best it could in VC1.

The only facility I have worked with in Europe, and continue to work with, is CMC/DMP. Even though they purchased from Sonic, I will still be supporting them. Same goes for MemoryTech in Japan, Deluxe, Technicolor, GDMX and 1k on the US.

But MemoryTech is not only authoring house, I understand that it sell some HD DVD authoring softwares (ask CMC, it use it, ?).
It develop this solutions from PureHD Ulead SDK.
So this company can not sell the VC1 solution encoder, because it's Sonic exclusivity ??!!!!


that you are going to use other codecs for projects?
In my work, we use only VC1 ( :mad: ), in europe it is little easy because we can rent the HD solution (encoder or authoring softwares).
I tried to explain that H264 is better but our clients want only VC1....
I hope to see in Europe a HD DVD in H264... (and I hope to see too TwinDisc, HD DVD on DVD)

8/28 and I wanted to ensure it looked the best it could in VC1.

So Heroes: Season One is using 1.07 then - I'm definitely looking forward to this one!

Stacy, are you satisfied with the results?

8/28 and I wanted to ensure it looked the best it could in VC1.
Man - that's a tough one - could be any of the following...

Blades of Glory (Paramount)
Dawn of the Dead (2004) (Universal)
Heroes: Season One (Universal)
The Hurricane (2000) (Universal)
Notting Hill (Universal)

Maybe the Paramount release, since they may have done a different encode for the BD version?

But I hope it's Dawn of the Dead, The Hurricane, or Notting Hill.

Stacy, are you satisfied with the results?

I have not seen the final results. I have heard positive things, but won't know for sure until my order comes in. If I want to see how something turned out, I have to buy or netflix it.

So this company can not sell the VC1 solution encoder

They can sale their own solution based on the Enterprise SDK. Eventually Sonic will have their own app, built on the SDK, as well and we will mostly likely, once again, be providing PEP on our own.

I expect PEP to always produce the best VC-1 results, at least as long as I am working on it.

Have you contaced Xavier about your issues? If not, I recommend it.

I hear ya. I used to be able to write batch files to encode 1000s of combinations and now I have to create xml files and I just can't do as much in the same time frame. vc1_enc.exe also never provided 3rd pass. (what we call re-encoding)

Yes, really better way for make test and like you say "1000s of combinations". Well we use vc1_enc.exe like that for multipass encoding:

@REM 6Mbps - 1080p24 - 3 pass insame profil

vc1_enc.exe -i C:\Master\1920x1080.yuv -o Temp\1080p_1.vc1 -w 1920 -h 1080 -framerate 29.97 -frametype progressive -telecine 1 -addeos -rate 6000 -maxrate 20000 -buffer 1843200 -gopperiod 14 -mvrange 1 -bframes 2 -bframeposopt 1 -bdeltaqp AdaptiveWeak -loopfilter 1 -deblocking 1 -complexity 0 -chromasearch 4 -motionmatch 2 -perceptual 2 -dquantstrength 0 -adaptivedeadzone 1 -favorinterlevel 3 -log Temp\stat_1.txt

vc1_enc.exe -i C:\Master\1920x1080.yuv -o Temp\1080p_2.vc1 -w 1920 -h 1080 -framerate 29.97 -frametype progressive -telecine 1 -addeos -rate 6000 -maxrate 20000 -buffer 1843200 -gopperiod 14 -mvrange 1 -bframes 2 -bframeposopt 1 -bdeltaqp AdaptiveWeak -loopfilter 1 -deblocking 1 -complexity 4 -chromasearch 4 -motionmatch 2 -perceptual 2 -dquantstrength 0 -adaptivedeadzone 1 -favorinterlevel 3 -2pass Temp\stat_1.txt -log Temp\stat_2.txt

vc1_enc.exe -i C:\Master\1920x1080.yuv -o Temp\1080p_3.vc1 -w 1920 -h 1080 -framerate 29.97 -frametype progressive -telecine 1 -addeos -rate 6000 -maxrate 20000 -buffer 1843200 -gopperiod 14 -mvrange 1 -bframes 2 -bframeposopt 1 -bdeltaqp AdaptiveWeak -loopfilter 1 -deblocking 1 -complexity 4 -chromasearch 4 -motionmatch 2 -perceptual 2 -dquantstrength 0 -adaptivedeadzone 1 -favorinterlevel 3 -2pass Temp\stat_2.txt

They can sale their own solution based on the Enterprise SDK. Eventually Sonic will have their own app, built on the SDK, as well and we will mostly likely, once again, be providing PEP on our own.

Sonic resell application from other companies:
CineVision "core coder" is from Mainconcept
CineVision PSE is from MS (no SDK or code gave at Sonic said Ben, MS do software Sonic sell it!)



I expect PEP to always produce the best VC-1 results, at least as long as I am working on it.

Agree, PEP is the best VC1 tool today.



Have you contaced Xavier about your issues? If not, I recommend it.

Who is Xavier ?

Hi all,

no news about VC1 encoded file from MS which can do better than H264.
Why ? Ms is in holiday (I hope), the sun shine in US (I hope too, because lot of
part in Europe we have only cloudy and rain).

I understand that we will have (post of Ben in Doom9 forum) 10min of true video
(first test used CG animation video, easy for all codec).
I'm sure that H264 with last grain optimization from some company (like VC1 optimization)
will do a nice encoding and " I " practically sure better than VC1 at 6Mbps.

I really want to see if PEP 1.07 (aka last version of CineVision PSE) can do better
( because with my test H264 encoder in CLI mode win and do better quality than
VC1 up to 40%: my test on my video with grain and without).

May be someone can ask Ateme company to particpate at this challenge.
Because in 2005, MSU H264 challenge declared Ateme the best H264 encoder
(better that x264 use in this challenge). I hope they can do, but I know
they work on UHD video, 7.680 × 4.320, with NHK, they will use H264 codec
(new profile ? or level ? will be create).




Regards,





Golgot13

no news about VC1 encoded file from MS which can do better than H264. Why ? Ms is in holiday (I hope), the sun shine in US (I hope too, because lot of part in Europe we have only cloudy and rain).
We're in the late stages of a couple of releases right now, but we should get to it over the next few weeks.

I understand that we will have (post of Ben in Doom9 forum) 10min of true video (first test used CG animation video, easy for all codec). I'm sure that H264 with last grain optimization from some company (like VC1 optimization) will do a nice encoding and " I " practically sure better than VC1 at 6Mbps.
We shall see...

I really want to see if PEP 1.07 (aka last version of CineVision PSE) can do better ( because with my test H264 encoder in CLI mode win and do better quality than VC1 up to 40%: my test on my video with grain and without).
Actually PEP 1.0.7 hasn't been released - it's the next big update (and probably should have been called 2.0 - it's our biggest update ever).

Actually PEP 1.0.7 hasn't been released - it's the next big update (and probably should have been called 2.0 - it's our biggest update ever).


There is lot of change on core 2.0 ?
The quality is increase, sure, but how many ?

This core 2.0 will be available only for CineVision PSE ?


I retest the compression with preset of Sagitaire (aka Zodiaque here),
I surprice by MPEG2 and VC1 level. Why on this configuration VC1 and MPEG2
are at same level ????? :eek:

The best MPEG2 encoder can make encoded file at same quality than PEP in VC1
(with specific bitrate: 6Mbps and 12Mbps).
I really hope that you will change this quickly, because if Sony come back
with BD encoded with this encoder (with average at 25Mbps), H264 and VC1 will be
dead for market development use (fortunately Warner use VC1 on HD DVD and BD,
some japaneses compagnies use H264 on HD DVD).


If some people can ask Toshiba, Sony, Panasonic, Thomson and CinemaCraft to test
their H264 encoder, it will be cold.
And I sure MS with PEP will win (there is lot of bad H264 implementation, not optimized)
against many of these encoders (in specific situations). But each encoder has his advantages
(eg Thomson and HD DVD FGT option; Sony and network encoding process;...). :D
And lot of people will surprise by this test.



Golgot13

There is lot of change on core 2.0?
The quality is increase, sure, but how many ?

This core 2.0 will be available only for CineVision PSE?
You'll see soon enough :).

I retest the compression with preset of Sagitaire (aka Zodiaque here),
I surprice by MPEG2 and VC1 level. Why on this configuration VC1 and MPEG2
are at same level ????? :eek:
Because the .vc1 encode isn't optimal, as previously discused, and isn't representative of the workflow or settings used in shipping titles.

You'll see soon enough :).


I hope, but the price of CineVision PSE will stop some company.
So we will have title encoded by this company which use old core
with a less quality video.





Because the .vc1 encode isn't optimal, as previously discused, and isn't representative of the workflow or settings used in shipping titles.

No, I made some encoding process like shipping title and it is same
(on CineVision PSE and PEP encoder). I tried to optimized some options
in encoder's GUI but no success.
Can you make the best encoding to verify with VC1 codec?





Golgot13

I retest the compression with preset of Sagitaire (aka Zodiaque here),
I surprice by MPEG2 and VC1 level. Why on this configuration VC1 and MPEG2
are at same level ????? :eek:

The best MPEG2 encoder can make encoded file at same quality than PEP in VC1
(with specific bitrate: 6Mbps and 12Mbps).


Got any screenshots?

Because the .vc1 encode isn't optimal, as previously discused, and isn't representative of the workflow or settings used in shipping titles.

Well now test with pep.exe at 6 Mbps with PP1 forced: 43.33 dB

Minimum Average Maximum

Y: Mean Absolute Deviation: 0.0000 13.5207 5.4359
U: Mean Absolute Deviation: 0.0000 0.8102 1.8150
V: Mean Absolute Deviation: 0.0000 0.8846 1.5686
Sum: Mean Absolute Deviation: 0.0000 25.6513 3.6997

Y: Mean Deviation: -0.9311 0.6090 1.0109
U: Mean Deviation: -0.3636 0.0775 0.5885
V: Mean Deviation: -0.5276 0.0748 0.3277
Sum: Mean Deviation: -0.6207 1.3430 0.6739

Y: Mean Square Error: 0.0000 4.1491 65.0187
U: Mean Square Error: 0.0000 0.6987 6.8956
V: Mean Square Error: 0.0000 0.7902 5.9206
Sum: Mean Square Error: 0.0000 3.0142 43.4682

Y: Root Mean Square Error: 0.0000 1.8002 8.0634
U: Root Mean Square Error: 0.0000 0.7003 2.6259
V: Root Mean Square Error: 0.0000 0.7548 2.4332
Sum: Root Mean Square Error: 0.0000 1.5514 6.5930

Y: PSNR: 30.0004 41.9513 1.#INF
U: PSNR: 39.7451 49.6881 1.#INF
V: PSNR: 40.4072 49.1536 1.#INF
Sum: PSNR: 31.7491 43.3391 1.#INF
Minimum Average Maximum

Mean Absolute Deviation: 0.0000 25.6513 3.6997
Mean Deviation: -0.6207 1.3430 0.6739
PSNR: 31.7491 43.3391 1.#INF


And now result with vc1_enc.exe with PP1 in flag: 43.75 dB

Minimum Average Maximum

Y: Mean Absolute Deviation: 0.0000 12.3687 5.4406
U: Mean Absolute Deviation: 0.0000 0.7497 1.5126
V: Mean Absolute Deviation: 0.0000 0.8131 1.4572
Sum: Mean Absolute Deviation: 0.0000 23.4965 3.6962

Y: Mean Deviation: -1.0000 0.3257 0.6041
U: Mean Deviation: -0.1913 0.0610 0.6475
V: Mean Deviation: -0.4243 0.0506 0.3930
Sum: Mean Deviation: -0.6667 0.7927 0.4057

Y: Mean Square Error: 0.0000 3.8036 67.1990
U: Mean Square Error: 0.0000 0.5776 4.8270
V: Mean Square Error: 0.0000 0.6528 4.0074
Sum: Mean Square Error: 0.0000 2.7408 44.9051

Y: Root Mean Square Error: 0.0000 1.6975 8.1975
U: Root Mean Square Error: 0.0000 0.6459 2.1970
V: Root Mean Square Error: 0.0000 0.6934 2.0019
Sum: Root Mean Square Error: 0.0000 1.4620 6.7011

Y: PSNR: 29.8572 42.3289 1.#INF
U: PSNR: 41.2941 50.5149 1.#INF
V: PSNR: 42.1022 49.9830 1.#INF
Sum: PSNR: 31.6078 43.7521 1.#INF
Minimum Average Maximum

Mean Absolute Deviation: 0.0000 23.4965 3.6962
Mean Deviation: -0.6667 0.7927 0.4057
PSNR: 31.6078 43.7521 1.#INF


And here the xml project for pep.exe:
- default RC and HVS setting from 1080p_HDDVD.PEP.xml template
- Complexity at maxi, Chroma at maxi, bframe at 7

<?xml version="1.0" encoding="utf-8"?>
<Project xmlns="http://schemas.microsoft.com/DigitalMedia/Codecs/2006/12/01/ParallelEncoderProject">
<ProjectName>Project1</ProjectName>
<PublishedFileURL>C:\Master\Project1.vc1</PublishedFileURL>
<PublishedLogFileURL>C:\Master\Project1.log</PublishedLogFileURL>
<OutputDirectory>C:\Master</OutputDirectory>
<SourceFiles>
<SourceFile>C:\Master\1920x1080.yuv</SourceFile>
</SourceFiles>
<AllowPtoISwitch>true</AllowPtoISwitch>
<AspectRatioIndex>1</AspectRatioIndex>
<BFrames>7</BFrames>
<Bitrate>6000</Bitrate>
<BufferSize>1843200</BufferSize>
<ChromaSearch>full</ChromaSearch>
<ClosedCaptionsPresent>false</ClosedCaptionsPresent>
<ColorFormatFlag>false</ColorFormatFlag>
<Denoise>false</Denoise>
<EncoderComplexity>4</EncoderComplexity>
<EncodingMode>VBR</EncodingMode>
<EncodingHeight>1080</EncodingHeight>
<EncodingWidth>1920</EncodingWidth>
<FrameRate>29.97</FrameRate>
<GOPLength>14</GOPLength>
<GOPType>Open</GOPType>
<Height>1080</Height>
<InterlaceMode>InterlacedTelecine</InterlaceMode>
<LoopFilter>true</LoopFilter>
<MarkIn>0</MarkIn>
<MarkOut>15689</MarkOut>
<MinEncodeQP>1.0</MinEncodeQP>
<MVRange>1</MVRange>
<PeakBitrate>20000</PeakBitrate>
<PDeltaQP>0.0</PDeltaQP>
<PerceptualOptionLevel>3</PerceptualOptionLevel>
<Segments>1</Segments>
<ThreadCount>0</ThreadCount>
<Width>1920</Width>
<BDeltaQP>
<Relative>0.0</Relative>
</BDeltaQP>
<FirstPassSettings>
<ChromaSearch>off</ChromaSearch>
<EncoderComplexity>0</EncoderComplexity>
<MVRange>0</MVRange>
</FirstPassSettings>
<CodecData>
<Property name="BFramePosOpt" type="long" value="1"/>
<Property name="DarkNoiseControl" type="long" value="0"/>
<Property name="DCBias" type="long" value="1"/>
<Property name="DQuantStrength" type="long" value="4"/>
<Property name="FavorInterLevel" type="long" value="3"/>
<Property name="MotionMatch" type="long" value="4"/>
</CodecData>
<UIData>
<RecentPublishedFileURL/>
</UIData>
</Project>


If you want really result from pep then I can use result from pep.
Anyway previous result from vc1_enc.exe with optimized setting are really better for PSNR/SSIM.
If the xml profil is not the best, it's possible to make other encoding with better xml profil.

Got any screenshots?

Encoding are available ... really better for subjective test.

Screenshots doesn't mean anything. With a "good" screenshots it's really simple to "prove" that MPEG2 is really better than H264. I will search and choose simply particular frame for each video codec: Iframe for MPEG2 and BFrame for H264.

7 B frames...why?

7 B frames...why?

See post of Ben on Doom9 forum:

http://forum.doom9.org/showthread.php?p=1029902#post1029902


The challenge interest some good H264 developper who say that VC1 can do better and can be optimized:
Today we will not really know if the standard VC1 or H264 is the better but that this implementation for this use is better....

I'm not agree with this point I want to hope at VC1 format, MS can show us it's true but MS don't show
any encoded file for this challenge...



Golgot13

You'll see soon enough :).
Because the .vc1 encode isn't optimal, as previously discused, and isn't representative of the workflow or settings used in shipping titles.


Ben, I really don't understand why you do show a best result with VC1 technology.
More than one week after with no encoding file from MS which show than VC1 is much better
than MPEG2 !!!!!

And Zodiaque/Sagitaire show VC1 encoded file from PEP use on professional workflow....
(not from me, because my free VC1 toolkit is lock)

Ben, I really don't understand why you do show a best result with VC1 technology.
More than one week after with no encoding file from MS which show than VC1 is much better
than MPEG2 !!!!!
Because our entire team is heads down prepping for several simultaneous releases. The only reason I have time to post here is when all my machines are busy rendering :).

And Zodiaque/Sagitaire show VC1 encoded file from PEP use on professional workflow....
(not from me, because my free VC1 toolkit is lock)
Again, it's not optimal.

1-2 B-frames would be better. Motion Search Range of 1 during 1st pass would be better.

7 B frames...why?

It's a good question because it's completely useless. Anyway you can use 7 bframes in vc1 specification for HDDVD. Anyway 7 adaptatives bframes or 2 adaptatives bframes don't change the result for this test.

Again, it's not optimal.

1-2 B-frames would be better. Motion Search Range of 1 during 1st pass would be better.


The result is same with you're optimal with PEP:
H264 is better than VC1 and the result of VC1 is near of MPEG2.

Read:
http://forum.doom9.org/showthread.php?p=1032840#post1032840



And about this result and the metric read this link to MS website:
http://www.microsoft.com/windows/windowsmedia/howto/articles/vc1techoverview.aspx#VC1ComparedtoOtherCodecs

MS used the equivalent metric to prouve last year that VC1 is better than MPEG2 and near of H264.



But I prefer that the encoding come directly from Ben and MS team.
So I wait the encoding file :D (near of 2 week without encoding file which show
than VC1 is really better than MPEG2 from best MPEG2 encoder!!!).



Ben, we wait you :p , I really sure you can do better result with PEP 1.07
(because 1.06 can not).



Regards,





Golgot 13

It's a good question because it's completely useless. Anyway you can use 7 bframes in vc1 specification for HDDVD. Anyway 7 adaptatives bframes or 2 adaptatives bframes don't change the result for this test.
Sure it will.

Also, as Zambelli said, it's NOT the max B-frames value, it's the average B-frame.

If you're trying to maximize PSNR, you want to use 1 B-frame.

Sure it will.

Also, as Zambelli said, it's NOT the max B-frames value, it's the average B-frame.

If you're trying to maximize PSNR, you want to use 1 B-frame.


Hi Ben,

I don't understand because Sagitairre, aka Zodiaque, use a preset from you given for this video and challenge ??????
So you gave a bad preset at Sagitaire or your preset is not the best (no time).

Last, Sagitairre made this encode with your preset to finish the challenge (I think it will be the "Summer Challenge")
because you can not give a encoding file for it (Sagitairre made encoding in one night, not two week), no time ?

I think it will nice to test PEP with professionnal H264 software like Elecard or MainConcept for Winter Challenge.
(there is some developper of this compagny on Doom9 ;) )

And I hope zambelli contact MSU (Moscow State University) VC1 participate at annual codec comparaison:
http://www.compression.ru/video/codec_comparison/index_en.html
We will know if VC1 is a good codec after this, and know if this link is true (?) or will need to updated :D
http://www.microsoft.com/windows/windowsmedia/howto/articles/vc1techoverview.aspx#VC1ComparedtoOtherCodecs



Regards





Golgot13

I never told him to use 7 B-frames, I just said that was a constraint not mandated by HD DVD. For Elephant's dream, most of it would use 0-2 B-frames optimally (credits might use more).

Ben,

Are there any plans to speed up encoding with PEP (CineVision PSE)?
It just simply to slow.
Will I be ever able to encode on a single workstation at eg. 10 FPS :)

Andrew

Are there any plans to speed up encoding with PEP (CineVision PSE)?
It just simply to slow.
Will I be ever able to encode on a single workstation at eg. 10 FPS :)

Sure, we're constantly working on performance optimization. Two big constant areas of work are optmization for new vector instructions (like SSE4) and taking advantage of multiple cores.

Of course, PEP itself stands for "Parallel Encoder Program" - the assumption is that it's running in a grid configuration. But grids are getting a lot smaller every year - I've seen some very impressive desktop-sized blade systems that can run PEP very nicely.

I never told him to use 7 B-frames, I just said that was a constraint not mandated by HD DVD. For Elephant's dream, most of it would use 0-2 B-frames optimally (credits might use more).


Hi Ben,

we answer at this ask on doom9 forum, and you can read some others here:
http://forum.doom9.org/showthread.php?p=1033441#post1033441


I want to know what VC1 can do really, to show at all the best of VC1 encoding.
Sagitairre know a person who can make the best encoding with PEP solution (or vc1_enc.exe ?).
He know well this encoder (not like me because I prefer H264 and my VC1 toolkit is lock)
he said there is more option on vc1_enc.exe, use by PEP.exe, than on PEP Gui (preset).

The challenge start many month ago (much to my mind, one year) but we could not finish it
because we don't have HD DVD specification (you never give all data of this) and encoder.
Today after one year, the first result with ED movie show that H264 is the best encoder, and
VC1 and MPEG have same result ( ???!!!! ). You said you can do better but after two week nothing (no file no result).


I hope PEP with 1.07 core can do better (may be you can ask at development to optimize it and
use the mask option of vc1_enc.exe to have best result). After that I will change my position between quality of H264 and VC1 codec.

I never told him to use 7 B-frames, I just said that was a constraint not mandated by HD DVD. For Elephant's dream, most of it would use 0-2 B-frames optimally (credits might use more).

Well here encoding with your optimal setting. Produce better OPSNR but not better SSIM.


Well now test with pep.exe at 6 Mbps with PP0: 44.10 dB for OPSNR and 85.72 for SSIM

Minimum Average Maximum

Y: Mean Absolute Deviation: 0.0000 13.1763 4.5640
U: Mean Absolute Deviation: 0.0000 0.7642 1.7663
V: Mean Absolute Deviation: 0.0000 0.8234 1.6109
Sum: Mean Absolute Deviation: 0.0000 24.8400 3.1151

Y: Mean Deviation: -1.0000 -0.5879 0.6439
U: Mean Deviation: -0.3941 0.0236 0.6183
V: Mean Deviation: -0.4885 0.0192 0.5467
Sum: Mean Deviation: -0.6667 -0.8277 0.3945

Y: Mean Square Error: 0.0000 3.4889 43.9233
U: Mean Square Error: 0.0000 0.5766 6.1960
V: Mean Square Error: 0.0000 0.6391 4.7024
Sum: Mean Square Error: 0.0000 2.5286 29.4009

Y: Root Mean Square Error: 0.0000 1.5742 6.6275
U: Root Mean Square Error: 0.0000 0.6369 2.4892
V: Root Mean Square Error: 0.0000 0.6793 2.1685
Sum: Root Mean Square Error: 0.0000 1.3636 5.4223

Y: PSNR: 31.7039 42.7039 1.#INF
U: PSNR: 40.2097 50.5222 1.#INF
V: PSNR: 41.4076 50.0752 1.#INF
Sum: PSNR: 33.4472 44.1020 1.#INF
Minimum Average Maximum

Mean Absolute Deviation: 0.0000 24.8400 3.1151
Mean Deviation: -0.6667 -0.8277 0.3945
PSNR: 33.4472 44.1020 1.#INF


And now result with vc1_enc.exe with PP1 in flag: 43.75 dB for OPSNR and 86.41 for SSIM

Minimum Average Maximum

Y: Mean Absolute Deviation: 0.0000 12.3687 5.4406
U: Mean Absolute Deviation: 0.0000 0.7497 1.5126
V: Mean Absolute Deviation: 0.0000 0.8131 1.4572
Sum: Mean Absolute Deviation: 0.0000 23.4965 3.6962

Y: Mean Deviation: -1.0000 0.3257 0.6041
U: Mean Deviation: -0.1913 0.0610 0.6475
V: Mean Deviation: -0.4243 0.0506 0.3930
Sum: Mean Deviation: -0.6667 0.7927 0.4057

Y: Mean Square Error: 0.0000 3.8036 67.1990
U: Mean Square Error: 0.0000 0.5776 4.8270
V: Mean Square Error: 0.0000 0.6528 4.0074
Sum: Mean Square Error: 0.0000 2.7408 44.9051

Y: Root Mean Square Error: 0.0000 1.6975 8.1975
U: Root Mean Square Error: 0.0000 0.6459 2.1970
V: Root Mean Square Error: 0.0000 0.6934 2.0019
Sum: Root Mean Square Error: 0.0000 1.4620 6.7011

Y: PSNR: 29.8572 42.3289 1.#INF
U: PSNR: 41.2941 50.5149 1.#INF
V: PSNR: 42.1022 49.9830 1.#INF
Sum: PSNR: 31.6078 43.7521 1.#INF
Minimum Average Maximum

Mean Absolute Deviation: 0.0000 23.4965 3.6962
Mean Deviation: -0.6667 0.7927 0.4057
PSNR: 31.6078 43.7521 1.#INF


And here the xml project for pep.exe:
- All HVS setting desactived for produce best PSNR.
- 1 thread for the encoding (curious way for P.E.P., isn't it?)
- Complexity at maxi, Chroma at maxi, bframe at 2.


<?xml version="1.0" encoding="utf-8"?>
<Project xmlns="http://schemas.microsoft.com/DigitalMedia/Codecs/2006/12/01/ParallelEncoderProject">
<ProjectName>Project1</ProjectName>
<PublishedFileURL>C:\Master\Project1.vc1</PublishedFileURL>
<PublishedLogFileURL>C:\Master\Project1.log</PublishedLogFileURL>
<OutputDirectory>C:\Master</OutputDirectory>
<SourceFiles>
<SourceFile>C:\Master\1920x1080.yuv</SourceFile>
</SourceFiles>
<AllowPtoISwitch>true</AllowPtoISwitch>
<AspectRatioIndex>1</AspectRatioIndex>
<BFrames>2</BFrames>
<Bitrate>6000</Bitrate>
<BufferSize>1843200</BufferSize>
<ChromaSearch>full</ChromaSearch>
<ClosedCaptionsPresent>false</ClosedCaptionsPresent>
<ColorFormatFlag>false</ColorFormatFlag>
<Denoise>false</Denoise>
<EncoderComplexity>4</EncoderComplexity>
<EncodingMode>VBR</EncodingMode>
<EncodingHeight>1080</EncodingHeight>
<EncodingWidth>1920</EncodingWidth>
<FrameRate>29.97</FrameRate>
<GOPLength>14</GOPLength>
<GOPType>Open</GOPType>
<Height>1080</Height>
<InterlaceMode>InterlacedTelecine</InterlaceMode>
<LoopFilter>true</LoopFilter>
<MarkIn>0</MarkIn>
<MarkOut>15689</MarkOut>
<MinEncodeQP>1.0</MinEncodeQP>
<MVRange>1</MVRange>
<PeakBitrate>20000</PeakBitrate>
<PDeltaQP>0.0</PDeltaQP>
<PerceptualOptionLevel>0</PerceptualOptionLevel>
<Segments>1</Segments>
<ThreadCount>1</ThreadCount>
<Width>1920</Width>
<BDeltaQP>
<Relative>0.0</Relative>
</BDeltaQP>
<FirstPassSettings>
<ChromaSearch>off</ChromaSearch>
<EncoderComplexity>2</EncoderComplexity>
<MVRange>1</MVRange>
</FirstPassSettings>
<CodecData>
<Property name="BFramePosOpt" type="long" value="1"/>
<Property name="DarkNoiseControl" type="long" value="0"/>
<Property name="DCBias" type="long" value="1"/>
<Property name="DQuantStrength" type="long" value="0"/>
<Property name="FavorInterLevel" type="long" value="3"/>
<Property name="MotionMatch" type="long" value="4"/>
</CodecData>
<UIData>
<RecentPublishedFileURL/>
</UIData>
</Project>

I want to know what VC1 can do really, to show at all the best of VC1 encoding.
Sagitairre know a person who can make the best encoding with PEP solution (or vc1_enc.exe ?).
He know well this encoder (not like me because I prefer H264 and my VC1 toolkit is lock)
he said there is more option on vc1_enc.exe, use by PEP.exe, than on PEP Gui (preset).


vc1_enc.exe and pep.exe use exactly the same core enginer with exactly the same command line and exactly the same result with same setting:
- No segment reencoding and paralleling support for vc1_enc.exe
- No deblocking option for pep.exe, anyway PP seem not supported by HDDVD. Really bad new for VC1 at low bitrate.
- No demo limitation for vc1_enc.exe, no heavy xml project for vc1_enc.exe, no heavy gui for vc1_enc.exe.

Well here encoding with your optimal setting.
I haven't communicated any optimal settings to you :). I've just been suggesting some obvious place for improvement. I've asked some of our PEP experts to take a look at the actual source and see what they can come up with.

In particular, I'm not confident than 2 B-frames is optimal for the entire clip. And I fear my mental translation from some of your XML properties is wrong (we've largely remapped the XML for the next version).

In particular, I'm not confident than 2 B-frames is optimal for the entire clip.


Yes, it's why the first test used the 2-B frames preset.


And I fear my mental translation from some of your XML properties is wrong (we've largely remapped the XML for the next version).

Ben, show us the possibility of VC1 codec :o :o

And I really want to test H264 on your best video for VC1 process.
I wait too. :p :p :p

I'll be getting to it.

I hate to plead lameness, but I work from home in Oregon, and my main encoding machine fried its power supply last week. So I'm driving up to Microsoft tonight to pick up a new machine to do this encode. I'll try (fingers crossed) to get it done before I leave for my Stanford class next week, vacation the week after, then IBC after that...

I know this sounds like a simple thing to do, but we're sprinting towards a number of IBC-related projects, so finding time to squeeze this in has been more of a challenge than you'd expect.

But we'll get it done somehow!

That said, I'm not sure why you set the bar at us beating H.264 here. Our story has always been that VC-1's biggest area of advantage is in preservation of film grain at moderate data rates. This kind of digital intermediate produced 3D source without any grain whatsoever plays to H.264's strengths, and isn't representative of the bulk of HDM content.

I'm also trying to find some good 35mm film source content that we can distribute source for, that can be used for more typical compression test scenarios. I'll be looking through a library of stuff tomorrow.

I'll be getting to it.

I hate to plead lameness, but I work from home in Oregon, and my main encoding machine fried its power supply last week. So I'm driving up to Microsoft tonight to pick up a new machine to do this encode. I'll try (fingers crossed) to get it done before I leave for my Stanford class next week, vacation the week after, then IBC after that...

I know this sounds like a simple thing to do, but we're sprinting towards a number of IBC-related projects, so finding time to squeeze this in has been more of a challenge than you'd expect.

But we'll get it done somehow!

That said, I'm not sure why you set the bar at us beating H.264 here. Our story has always been that VC-1's biggest area of advantage is in preservation of film grain at moderate data rates. This kind of digital intermediate produced 3D source without any grain whatsoever plays to H.264's strengths, and isn't representative of the bulk of HDM content.

I'm also trying to find some good 35mm film source content that we can distribute source for, that can be used for more typical compression test scenarios. I'll be looking through a library of stuff tomorrow.
Ben, I'm following this little contest for a while now. Thanks to you and the other guys for the effort and contribution. However, I never understood why the decision was made to use a full synthetic source (which doesn't show film grain, noise, scratches, speckles, natural light and surface) and not transferred analog film in the first step. I got the feeling that most of the discussions about decent or "bad" picture quality on this board are about analog transfers. Synthetic sources will always look synthetic regardless if the PSNR is 44.1 or 43.75, at least from the standpoint of the "average" HT fan. As we're here for natural movies, and preserving film, I hope to see a contest based on film material.

Ben, I'm following this little contest for a while now. Thanks to you and the other guys for the effort and contribution. However, I never understood why the decision was made to use a full synthetic source (which doesn't show film grain, noise, scratches, speckles, natural light and surface)


Simple because it's free video, we can use it. (I don't think that some major can give us some trailer for this :( )



and not transferred analog film in the first step. I got the feeling that most of the discussions about decent or "bad" picture quality on this board are about analog transfers.


If you can give some video like Ben to finish the challenge, you're wellcome.



Synthetic sources will always look synthetic regardless if the PSNR is 44.1 or 43.75, at least from the standpoint of the "average" HT fan. As we're here for natural movies, and preserving film, I hope to see a contest based on film material.

I hope too, we test the HD DVD and BD codec on movie and we show the result (we don't put some result on website without possibilty
to prove what we say, eg MS website comparison codec...).

But I test in my work H264 and VC1 on movie with grain and I have the same result H264 is better (with my video).
But all our client choose VC1 because there is lot of VC1 HD DVD Title....



And about HD DVD, I see you have a european HD DVD player: I angry about Toshiba which don't give a firmware update to play 50Hz (25fps) video!!!!!
I really think that HD DVD is for only for USA country (not Japan because no people buy it), I can say thank you (if MS was french :D all HD DVD player could play 50Hz....). :mad:

And about HD DVD, I see you have a european HD DVD player: I angry about Toshiba which don't give a firmware update to play 50Hz (25fps) video!!!!!
I'm European and I don't care about 50Hz support in the HD DVD players. Why? Simply because every content I'm interested in is native 24fps.

I'm European and I don't care about 50Hz support in the HD DVD players. Why? Simply because every content I'm interested in is native 24fps.

Fortunately for movie, but all concert live, all documentary, TV serial and TV movie are in 50Hz.
There is much more 50Hz program than movie....

HD DVD was the first on market (except old BD in Japan) but it is the late about 50Hz (BD support it ) ?????

Fortunately for movie, but all concert live, all documentary, TV serial and TV movie are in 50Hz.
There is much more 50Hz program than movie....
You are probably right. Personally, I'm not interested in live concerts, documentaries and TV movies. And the TV series I'm interested in are all 60i or 24p. So no 50Hz need for me. But I understand that that's just me. As far as I've heard, Toshiba was planning to add 50Hz support with a later firmware. Don't know why they didn't add it yet. But actually I'm happy about that cause this forces all studios to master in 24p. Otherwise some stupid European studios might have decided to continue applying PAL speedup to native 24fps movies. Anyway, I'm OT here, I guess... :o

I'll be getting to it. ...

I'm also trying to find some good 35mm film source content that we can distribute source for, that can be used for more typical compression test scenarios. I'll be looking through a library of stuff tomorrow.

Let's issue this as a challenge to the Madison Avenue guys to create the ultimate product placement short! Everybody here sing the praises of having an eternal commercial that gets played over and over on demos and tests everywhere. All they have to do is create a good clean 35mm film scan of a few minutes of whatever they want and place it in the public domain for our purposes. Pretty girls would be optional (desired) but they should choose things with a long shelf life. This thing would be circulating around for years (remember Lena?) if it wasn't legally encumbered.

Anybody know any advertising guys?

- Tom

Ben, I'm following this little contest for a while now. Thanks to you and the other guys for the effort and contribution. However, I never understood why the decision was made to use a full synthetic source (which doesn't show film grain, noise, scratches, speckles, natural light and surface) and not transferred analog film in the first step. I got the feeling that most of the discussions about decent or "bad" picture quality on this board are about analog transfers. Synthetic sources will always look synthetic regardless if the PSNR is 44.1 or 43.75, at least from the standpoint of the "average" HT fan. As we're here for natural movies, and preserving film, I hope to see a contest based on film material.
We'll it's not like there was a scientific process to choose the clip - it was merely the best readily available :).

I'm actually going to be picking up some 35mm source in the next few days that might work for a more movie-relevant test.

Also, I think the focus on the objective measurements, particularly PSNR, isn't accurate to real quality anymore. Based on the discussions, it sounds like x264 developmenti is very much driven by optimizing around PSNR and SSIM, while we have a staff of content reviewers that look at the stuff all day. There's plenty of features in our codec that we know improve subjective quality, but lower the objective measurements.

So, arguing about the numbers won't be nearly as interesting or relevant as arguing about how the actual clips work.

So, arguing about the numbers won't be nearly as interesting or relevant as arguing about how the actual clips work.


Hi Ben,

Thank you to choose the best video you can for the challenge. I hope it will possible to use this video on other challenge
(I think the best it's to put Microsoft logo inside, may be it will come a reference to test some codec like some video on MSU website).

About number, on your comparison codec website you (MS) give some number (?) to judge the codec efficiency.
And about the HVS, yes it's necessary when the PSNR difference between two different codec is no more than 0.7dB.
But if the difference is high than 0.7dB, the subjective view is no necessary because the quality is really better on high PSNR video.

About number, on your comparison codec website you (MS) give some number (?) to judge the codec efficiency.
And about the HVS, yes it's necessary when the PSNR difference between two different codec is no more than 0.7dB.
But if the difference is high than 0.7dB, the subjective view is no necessary because the quality is really better on high PSNR video.
But were do you get the 0.7 number from withou testng emperically? And retesting whenever perceptual algorithms change?

And again, PSNR and SSIM assume a perceptually uniform gamma, which isn't true for many or most modern displays. We've tuned our VC-1 implemntation to address those quality limitations, based on tons of feedback from the studios, so we wind up spending a lot more bits than other codecs in that low luma range. While this might not pay off in objective measurements, it sure makes the studio compresionists and consumers happy!

But were do you get the 0.7 number from withou testng emperically? And retesting whenever perceptual algorithms change?


From my teacher (who work on MPEG commitee). :D

Besides, if we have actual encoding results we can look at both the numbers, the clips themselves, and the opinions of panels of reviewers and form our own opinions. No reason we cannot personally weigh all three.

- Tom

I'm European and I don't care about 50Hz support in the HD DVD players. Why? Simply because every content I'm interested in is native 24fps.

You don't have native 24Fps - is slowed 23,976xxxxxx P0S with jitter.
This P0S of course a little "better" than speeded 25Fps P0S from film-celluloid (24Fps) source.

But my need in HD optical media (HDDVD or BD) is archiving w/o recompression FAMILY HD(V) 50Hz VIDEO:
Weddings, Born of child, Birthdays of parents and friends...
This matter for me, not Hollywood's 99% of upscaled s##t :D

From my teacher (who work on MPEG commitee). :D
And that metric probably works reasonably well in comparing one implementation of the same codec with another, assuming perceptually uniform gamma.

But it's just a rule of thumb, and there certainly willl be cases where it doesn't apply.

But it's just a rule of thumb, and there certainly willl be cases where it doesn't apply.

Yes, you're right after read my (old) note lesson from this teacher.
But there is a limit when the eye, subjective view, will say everytime this picture from this codec is better.

H264 and VC1 are two codec for HD video use. They have some advantages and disadvantages.
I really surprise by H264 codec (by some nice implementation).

Today the best choice is VC1, because this codec is optimized to be play on PC and it don't need a big CPU use.
But in 2008 or 2009, I will prefer H264 and I think it will be the codec support of HD DVD title
(Intel CPU and AMD COU with 2 core are enough)

...
Today the best choice is VC1, because this codec is optimized to be play on PC and it don't need a big CPU use.
But in 2008 or 2009, I will prefer H264 and I think it will be the codec support of HD DVD title
(Intel CPU and AMD COU with 2 core are enough)

I will tentatively go along with this also, unless we see other tests with a good HD film scan showing something different.

- Tom

Today the best choice is VC1, because this codec is optimized to be play on PC and it don't need a big CPU use.
I do not agree, anymore. With my current graphics card h264 acceleration works so well that h264 stresses my CPU less than even MPEG2!! Unfortunately ATI's latest cards (2400/2600) and Microsoft's VC-1 decoder don't like each other (no hardware acceleration at all!). Because of that VC-1 is not my preferred codec, anymore.

I do not agree, anymore. With my current graphics card h264 acceleration works so well that h264 stresses my CPU less than even MPEG2!!

Yes, because you have a new graphic card (no more than 1.5 years).
But lot of people have old PC (no Dual Core, no PCI Express Graphic Card,...).
If you have not a H264 acceleration graphic card (or best H264 directshow filter: CoreAVC),
you will need at minimum a P4 at 3.2Ghz to play H264 HD video (with some freeze). But for VC1 HD video,
it work on P4 2.4GHz (no freeze but the CPU is used at 100%).


Unfortunately ATI's latest cards (2400/2600) and Microsoft's VC-1 decoder don't like each other (no hardware acceleration at all!)
This is a little conflict that AMD will solve soon (because GPU will accelerate VC1 and H264 decoding).


I like your answer because you understand that CPU power use it's not a advantage of VC1 on new hardware.

This is a little conflict that AMD will solve soon (because GPU will accelerate VC1 and H264 decoding).
Are you sure about that?

The latest AMD/ATI cards support hardware acceleration mode D for VC-1, but not modes A, B and C. Unfortunately the MS VC-1 decoder only supports modes A and B (and in Vista C) but not mode D. Because of that there's zero hardware acceleration with the latest ATI cards and the MS VC-1 decoder.

But for VC1 HD video, it work on P4 2.4GHz (no freeze but the CPU is used at 100%).
Have you tried the movie "Deja Vu"? There are some scenes which are not fluid on my Core 2 Duo. Most other movies are fine. But I think we'll see more high bitrate VC-1 content sooner or later (e.g. from Disney). So P4 2.4GHz may not be enough for VC-1 soon, anymore.

Are you sure about that?

The latest AMD/ATI cards support hardware acceleration mode D for VC-1, but not modes A, B and C. Unfortunately the MS VC-1 decoder only supports modes A and B (and in Vista C) but not mode D. Because of that there's zero hardware acceleration with the latest ATI cards and the MS VC-1 decoder.


It will be update on new software HD DVD or BD player (Do you try HD DVD player on Nero ?).



Have you tried the movie "Deja Vu"? There are some scenes which are not fluid on my Core 2 Duo. Most other movies are fine. But I think we'll see more high bitrate VC-1 content sooner or later (e.g. from Disney). So P4 2.4GHz may not be enough for VC-1 soon, anymore.


For lot of VC1 HD video, P4 2.4GHz is enough, but all H264 HD with CABAC option can not be play on P4 2.4GHz
(without GPU acceleration). And if you use MBAFF + CABAC + HP@L4.1 + Cutsom Matrix, P4 at 3.4GHz is limit to play the video.
VC1 is easier to be played on more CPU (because it don't need lot of CPU power and beacuse VC1 decoder is optimized, thx MS).

[...]
And about HD DVD, I see you have a european HD DVD player: I angry about Toshiba which don't give a firmware update to play 50Hz (25fps) video!!!!!
I really think that HD DVD is for only for USA country (not Japan because no people buy it), I can say thank you (if MS was french :D all HD DVD player could play 50Hz....). :mad:I don't know why you brought up this off-topic issue in a codec comparison thread.

However...

We'll it's not like there was a scientific process to choose the clip - it was merely the best readily available :).

I'm actually going to be picking up some 35mm source in the next few days that might work for a more movie-relevant test.

Also, I think the focus on the objective measurements, particularly PSNR, isn't accurate to real quality anymore. Based on the discussions, it sounds like x264 developmenti is very much driven by optimizing around PSNR and SSIM, while we have a staff of content reviewers that look at the stuff all day. There's plenty of features in our codec that we know improve subjective quality, but lower the objective measurements.

So, arguing about the numbers won't be nearly as interesting or relevant as arguing about how the actual clips work.Ben, and Golgot13,

thanks for your answers. I now understand why this clip was chosen. I really hope you can dig up a natural material, Ben. Just for you to understand why I feel this is so important. As you might have noticed by my sig, I own a japanese Muse HiVision system. This is an analog compression system based on nyquist subband encoding. I have seen Muse Demo LDs which can easily hold up viewing tests with MPEG2 HD material (despite the fact that SNR is much worse on Muse, and bandwidth is cut at 27MHz I think), but with analog film sources it shows its issues and flaws.
So what would I expect from an "attempt to find an objective AVC vs. VC1 benchmark" - as the thread title suggests? I'd expect to see a superiority of the winning candidate by viewing encoded clips. In other words, numbers are numbers, but one has to see/notice a difference as well.

cheers, Torsten

It will be update on new software HD DVD or BD player (Do you try HD DVD player on Nero ?).
That doesn't help at all. I don't want to use PowerDVD nor Nero ShowTime. They both suck big time. I want to use MPC or ZoomPlayer. And the MS VC-1 decoder is the only VC-1 decoder which works well with MPC and ZoomPlayer. But as I said before:

MS VC-1 decoder + latest ATI cards = zero hardware acceleration

So what would I expect from an "attempt to find an objective AVC vs. VC1 benchmark" - as the thread title suggests? I'd expect to see a superiority of the winning candidate by viewing encoded clips. In other words, numbers are numbers, but one has to see/notice a difference as well.


Agree with you, but the PSNR difference between VC1 and H264 is High and it will see directly on video quality (for ED video on this test).
I wait to see the result about new video from Ben, and judge if VC1 is really better than H264 on movie (with grain).

judge if VC1 is really better than H264 on movie (with grain).
Seems to me........... if my memory is correct, that Paramount always favored VC-1 on their HD DVD releases.
Recently, I noted that Shooter came out with an AVC MPEG-4 encode on the HD DVD edition disc, then it looks like Paramount decided to use matching AVC MPEG-4 encodes on both the HD DVD and Blu-ray discs for Disturbia.

Are these two events serendipitous anomalies?
Or do they represent blips on the radar screen to be observed for a definite trend in the future, namely the increasing usage of AVC with their HD DVD editions or even eventual abandonment of VC-1 altogether ?

Are these two events serendipitous anomalies?
Or do they represent blips on the radar screen to be observed for a definite trend in the future, namely the increasing usage of AVC with their HD DVD editions or even eventual abandonment of VC-1 altogether ?

It's Paramount but Universal use and will use only VC1.
It's only politic not really because the quality is better... :(

Are there any plans to speed up encoding with PEP (CineVision PSE)?
It just simply to slow.

Agree with you. And it support only one core.



Will I be ever able to encode on a single workstation at eg. 10 FPS :)


Yes, no optimization yet, may be it will available on PEP 1.07



Zodiaque can you make a ED movie with grain (there is some filter to put grain or noise on video).
I think it will be the better way to have, quickly, a video with grain to finish the test of all codec. ;)

The difference in a film based shoot out is probably that there is very little information in the higher frequencies on film besides grain and other semi-random noise. Meanwhile CGI can create video with a basically flat MTF curve if it really wants. It is possible some codecs have been optimized around film properties. The AVC standard even includes FGT options to filter out the very highest frequencies and then supply the noise/grain with similar properties at display time (something I actually proposed on this forum some time before they announced it).

I don't know if anyone is using FGT but it does show some things may be optimized for a film like MTF curve. Assumptions of motion blur is another contributor to this curve for film since you can assume many hard to compress motion scenes will also have been filtered a bit by motion blur and you can get away with a bit more without being obvious.

- Tom

Agree with you. And it support only one core.
PEP has always been 4-way threaded!

Agree with you. And it support only one core.

PEP supports up to 4 HW threads (cores) per worker. If you had an 8-way machine, you would simply run two workers on the single machine along with the controller.

1.07 is actually slower than 1.06.

We have measured, on the best HW, less than 2x real-time encoding. Sony's MPEG2 HD encoder is 2x real-time. It is real-time per-pass and takes two passes. The <2x was on expensive HW, but still much less than the MEI AVC encoder and cheaper than the Sony AVC encoder. It does make it more than the Cinema Craft encoder.

I finally got around to looking at the x264 encode of ED. (12 Mbps version) and it looks great! None of the AVC encoders used for HD DVD or BD are as good as the x264 encoder. If I had to rate them in order of quality it would go:

Thompson
MEI
Sony
Cinema Craft
Toshiba

Both Paramount and Universal have left the choice up to the compressionist w/r/t which codec to use for a given title. They really just want the best looking title they can get. Sadly, things are now date driven, so titles have to be done quickly. HDM is coming out of R&D and now people are wanting to make some money, so they can't afford to tweak until it is perfect anymore.

I finally got around to looking at the x264 encode of ED. (12 Mbps version) and it looks great! None of the AVC encoders used for HD DVD or BD are as good as the x264 encoder.

This is what I like to hear :) Hopefully the big wigs take notice of this and improve their implementations because a bunch of enthusiasts are doing such a good job of it.

...
I finally got around to looking at the x264 encode of ED. (12 Mbps version) and it looks great! None of the AVC encoders used for HD DVD or BD are as good as the x264 encoder. ...

Yep. That may be the source of some confusion in the codec wars right now. I don't suppose there is any way to master commercial BD/HD DVD discs using X264, is there?

- Tom

I finally got around to looking at the x264 encode of ED. (12 Mbps version) and it looks great!

:D
Thanks the VIA team (they work too on VLC)


None of the AVC encoders used for HD DVD or BD are as good as the x264 encoder. If I had to rate them in order of quality it would go:

Thompson
MEI
Sony


What version of Thomson did you test?
Because I tested last year (september 2006).

And what is it MEI ?
-> Matsushita Encoder (Panasonic encoder) ?


Can you say more about Film Grain option (for Ben) ?


Regards,

It's Paramount but Universal use and will use only VC1.
It's only politic not really because the quality is better... :(

I agree. Let’s hope that Paramount stays “scientific” and does not let “politics” influence them in any way and continues their current course of action.

No further comment.

I finally got around to looking at the x264 encode of ED. (12 Mbps version) and it looks great! None of the AVC encoders used for HD DVD or BD are as good as the x264 encoder. If I had to rate them in order of quality it would go:

Thompson
MEI
Sony
Cinema Craft
Toshiba

Both Paramount and Universal have left the choice up to the compressionist w/r/t which codec to use for a given title. They really just want the best looking title they can get. Sadly, things are now date driven, so titles have to be done quickly. HDM is coming out of R&D and now people are wanting to make some money, so they can't afford to tweak until it is perfect anymore.

Well I make personaly test with other H264 implementation: ateme (nero) with core 16/12/2005 version and Mainconcept/Elecard with core 11/10/2006. These H264 implementation produce really close metric result but have really more advanced HVS function than x264 particulary for adaptative quantisation.

x264 is a very optimized H264 implementation for metric (simply the best codec in the world I think) but have some difficulty for preserve noise/grain. I make test with really noisy/grainy source and Elecard/Mainconcept/Ateme implementation is really a better choice if you want grain preservation with H264 (test at 8.5 Mbps with 1080p24 uncompressed source master).

IMO Thompson, ME, Sony, Cinema Craft and Toshiba implementation are simply optimized for high speed in high bitrate situation (BD50 scenario) and produce bad result at low/medium bitrate.

No optimal setting? All the ME/RDO search at maximum quality (complexity at maxi, chromasearch at maxi, SADT motionmatch) with default setting for DQuant and optimized setting for SSIM boost.
Actually, our latest PSNR/SSIM measurements show that encoding without dquant yields higher PSNR and SSIM scores.

This makes sense, actually. If you consider dquant as a way of shifting bits from highly textured areas to finer, smoother areas of the picture, this means that the emphasis might be on better preserving a minority of the picture at the expense of the majority of the picture. The more bits are spent on getting a shadow or gradient looking good, the less bits become available for the rest of the picture - and ultimately the rest of the frames within the buffer. This would adversely affect PSNR and SSIM (which are tuned to reflect similarity with the source), yet perceptually the dquant-encoder frames might be more pleasing because they wouldn't contain obvious banding or blocking in smooth areas. Sure, they'd also be less sharp in textured areas due to higher quantization, but keep in mind that typically lack of sharpness needs a point of reference to be noticed, whereas blocking or banding doesn't need any reference - it's always very obvious when it's there, even if you've never seen the source.

Come on ben it's the setting that you use yourself for this source with WMV9 AP implementation.
Neither Ben nor I ever recommend settings based on scoring highest SSIM scores. If we recommend dquant, for example, we do it because we think it's suitable for a particular scenario.

Actually, our latest PSNR/SSIM measurements show that encoding without dquant yields higher PSNR and SSIM scores.

This makes sense, actually. If you consider dquant as a way of shifting bits from highly textured areas to finer, smoother areas of the picture, this means that the emphasis might be on better preserving a minority of the picture at the expense of the majority of the picture. The more bits are spent on getting a shadow or gradient looking good, the less bits become available for the rest of the picture - and ultimately the rest of the frames within the buffer. This would adversely affect PSNR and SSIM (which are tuned to reflect similarity with the source), yet perceptually the dquant-encoder frames might be more pleasing because they wouldn't contain obvious banding or blocking in smooth areas. Sure, they'd also be less sharp in textured areas due to higher quantization, but keep in mind that typically lack of sharpness needs a point of reference to be noticed, whereas blocking or banding doesn't need any reference - it's always very obvious when it's there, even if you've never seen the source.



http://forum.doom9.org/showthread.php?p=1032840#post1032840

I make test with DQuant desactived and this setting improve the OPSNR but not the SSIM. In fact for OPSNR the most important improvement for pep v 1.0.7 seem to be the Rate Control. In my result with 1.0.6 the SSIM is really good (92.44 for 1.0.6 vs 92.70 for 1.0.7) and "bad" OPSNR (46.71 dB for 1.0.6 and 47.62 dB for 1.0.7). It's typically a Rate Control "bug" for v 1.0.6 because:

- Adaptative Quantisation in general case don't improve OPSNR and improve SSIM. Anyway in general case the delta for metric with and without AQ are very small.
- constant APSNR/SSIM with big OPSNR improvement mean always better Rate Control Optimisation.

I report at Benwaggoner a very fine Rate Control Analyse for pep v 1.0.6:
- pep v 1.0.6 don't use really the buffer. There are never buffer saturation even for really complex part.
- pep v 1.0.6 produce very bad frame in complex part with very low local PSNR in these part. Very bad frame produce bad OPSNR. Better OPSNR mean always more constant quality.

I think that I can produce even better result if I use part by part encoding for reduce the Rate Control "bug" in v 1.0.6. Anyway don't expect better result than H264 ... VC1 can't fight with H264 here simply because H264 produce better result in part by part comparison for 99% of the frame and Rate Control optimisation can't never change that. New VC1 encoding in progress ...

- Adaptative Quantisation in general case don't improve OPSNR and improve SSIM. Anyway in general case the delta for metric with and without AQ are very small.
In PEP 1.0.7 Dquant is more aggressive than in PEP 1.0.6 - that's why the SSIM difference we noticed between Dquant on/off was much bigger.

I did some tests with the v11 WMV9 encoder too, and was able to get SSIM up to 92.37 even with DQ. In fact, in WMV9 the SSIM was slightly higher with DQ than without it. It essentially boils down to how aggressively Dquant is moving bits from the textured regions to the smooth regions. If we assume that smooth regions represent the minority of an image area, pumping more bits into those regions will hurt the frame PSNR & SSIM more. So I'm guessing that for PSNR/SSIM some dquant is good, but once you go past the "sweet spot" - the metrics start to decline, and I think that's what we're seeing here.

- constant APSNR/SSIM with big OPSNR improvement mean always better Rate Control Optimisation.
This seems like a very interesting correlation. Can you elaborate on the second statement? Are you saying that if encode A and encode B have the same SSIM score, but encode B has a higher PSNR score - that's an indicator that encode B has better ratecontrol?

I report at Benwaggoner a very fine Rate Control Analyse for pep v 1.0.6:
- pep v 1.0.6 don't use really the buffer. There are never buffer saturation even for really complex part.
- pep v 1.0.6 produce very bad frame in complex part with very low local PSNR in these part. Very bad frame produce bad OPSNR. Better OPSNR mean always more constant quality.
That's very useful feedback. We very much appreciate that kind of feedback, thank you!
I think that I can produce even better result if I use part by part encoding for reduce the Rate Control "bug" in v 1.0.6. Anyway don't expect better result than H264 ... VC1 can't fight with H264 here simply because H264 produce better result in part by part comparison for 99% of the frame and Rate Control optimisation can't never change that. New VC1 encoding in progress ...
I honestly don't have any expectations - I'm more interested in this comparison from a purely engineering point of view. This type of research is always good data, regardless of whether it's favorable or not.

I agree. Let’s hope that Paramount stays “scientific” and does not let “politics” influence them in any way and continues their current course of action.

No further comment.

I come back from holiday.
Paramount announced that lot of people knew (rumour since May 2007) they will support only HD DVD.
Now it is not a "politic" influence but a "business/money" influence (150M). MS and Toshiba will give a team (?)
at Paramount to make HD DVD title for Free/gratis... ( EU commision will not like this...)
I sure there was same agreement with other companies (on HD DDVD and BD association)....
No consideration for final user, consumers :mad::mad::(


There are not "scientific" influence today :(
(about TV broadcaster in europe, it's a economic influence to choose HD H264).

14.08.07:
I'm actually going to be picking up some 35mm source in the next few days that might work for a more movie-relevant test.

For "few", 2 weeks is enough ? :p

I have the raw source now, but no chance I'll be able to do any work on it until after IBC.

I have the raw source now, but no chance I'll be able to do any work on it until after IBC.

You will go at IFA too ?
Give the access at the source I will try to do a nice encoding to IFA (31/08) and better for IBC (07/09).
And if you want to test the file on specific hardware player (not mux on HD DVD) give me the name of chipset
I will optimize the HVS for this chipset ;)

I will, may be, see you on this show (sure if there is nice girl on MS stand...) :D

It's a good half TB of source, which is why I haven't been able to do much with it :).

Now it is not a "politic" influence but a "business/money" influence (150M). MS and Toshiba will give a team (?)
at Paramount to make HD DVD title for Free/gratis... ( EU commision will not like this...)
That's pure speculation. Where did you read about Toshiba/MS giving a team of compressionists to Paramount?

Furthermore, there's this quote:
http://www.nytimes.com/2007/08/21/technology/21disney.html?_r=2&em&ex=1187841600&en=5ddd87161117ed10&ei=5087%0A&oref=slogin&oref=slogin
Paramount and DreamWorks Animation declined to comment. Microsoft, the most prominent technology company supporting HD DVDs, said it could not rule out payment but said it wrote no checks. “We provided no financial incentives to Paramount or DreamWorks whatsoever,” said Amir Majidimehr, the head of Microsoft’s consumer media technology group."

It's a good half TB of source, which is why I haven't been able to do much with it :).

1/2 TB of source fits on a $100 drive these days. If I mail you one can I have a copy? :)

- Tom

That's pure speculation. Where did you read about Toshiba/MS giving a team of compressionists to Paramount?

I want to say that MS will give a "nice" support and will make some encoding...
(Miami Vice was encoded by "nice" people)


Furthermore, there's this quote:
http://www.nytimes.com/2007/08/21/technology/21disney.html?_r=2&em&ex=1187841600&en=5ddd87161117ed10&ei=5087%0A&oref=slogin&oref=slogin

Yes, I know but there is more (eg StudioCanal and HD DVD: StudioCanal is not HD DVD prg member, ?, ...).:p

It's a good half TB of source, which is why I haven't been able to do much with it :).


It's near of a complete movie (500GB!!!).
For challenge, 10min to 30min is enough (I think 10min is really good).
You can use a lossless codec to safe lot of space (20-30GB) ....
And you will be able to do much after... :p

Soon x264 will do better encoding with "movie noise" (no FGM yet, but may be at the end of school year: for student project).

IMO Thompson, ME, Sony, Cinema Craft and Toshiba implementation are simply optimized for high speed in high bitrate situation (BD50 scenario) and produce bad result at low/medium bitrate.

Which is exactly how it should be :)

Andrew

Which is exactly how it should be :)



Not agree, now the last CPU power (with Bi Intel Quad) is enough to develop the picture optimization at medium and low bitrate.
Last development of H264 HD broadcaster encoder show it possible to have the speed (real time) and the quality (at low bitrate 4 to 9 Mbps).

You're in europe (?) you can see SkyHD TV (clear: test of new channel) and BBC HD.

Not agree, now the last CPU power (with Bi Intel Quad) is enough to develop the picture optimization at medium and low bitrate.
Last development of H264 HD broadcaster encoder show it possible to have the speed (real time) and the quality (at low bitrate 4 to 9 Mbps).

You're in europe (?) you can see SkyHD TV (clear: test of new channel) and BBC HD.

The two bolded statements have nothing at all to do with each other.

Sky HD broadcasts at rates between 10 and 20 mbps avergae bit rate (1920x1080i50 PAFF), BBC-HD broadcasts at roughly 19.5 mbps average bit rate (1440x1080i50 MBAFF).

Sky's 20 Mbps broadcasts generally look good, but do not reach the quality Premiere HD reaches at the same bitrate (different encoder settings). BBC-HD also looks good (given the advantage the lower horizontal resolution provides it hardly comes as a surprise).
Sky HD at 10-14 Mbps looks pretty bad.

So, where are these quality broadcasts at 4-9 mbps that you speak of? Certainly not on the channels you want people to look at.

Not agree, now the last CPU power (with Bi Intel Quad) is enough to develop the picture optimization at medium and low bitrate.
Last development of H264 HD broadcaster encoder show it possible to have the speed (real time) and the quality (at low bitrate 4 to 9 Mbps).

You're in europe (?) you can see SkyHD TV (clear: test of new channel) and BBC HD.

There is no point to optimise studio encoders for 4-10 Mbps. These encoders are used with 12 + Mbps and that the bitrate range what they should be optimise for. Broadcaster encoder is a totally different story.

Andrew

Sky HD broadcasts at rates between 10 and 20 mbps avergae bit rate (1920x1080i50 PAFF), BBC-HD broadcasts at roughly 19.5 mbps average bit rate (1440x1080i50 MBAFF).

Yes, but you will see soon the bitrate will decrease, because the encoder will be optimized
(saw demo at last NAB, see demo at IBC soon).


Sky's 20 Mbps broadcasts generally look good, but do not reach the quality Premiere HD reaches at the same bitrate (different encoder settings). BBC-HD also looks good (given the advantage the lower horizontal resolution provides it hardly comes as a surprise).
Sky HD at 10-14 Mbps looks pretty bad.

Agree, it was old implementation of Tandberg encoder (it was expensive, it will work some day more)


So, where are these quality broadcasts at 4-9 mbps that you speak of? Certainly not on the channels you want people to look at.

If you saw my post on doom9, yo will understand that I see and speak on specific tests:
- Like test of "Tour de France" on DVB-T in France with 3 differents and last H264 HD encoder (nice picture of Ateme encoder ;) )
- Or HD Astra demo channel (trailers of differents channel) (if you have a ftp I can upload some TS files at 8.5Mbps).
- Or HD IPTV in France (some tests in differents resolutions).

The best is to go and see demo at IBC of GrassValley (Thomson), Ateme, Harmonic (Electra7000), TandBerg,... encoder
All demo start at 7 or 8 Mbps. And all broadcaster will use 8 or 9Mbps to broadcast their channel
because the bandwidth is expensive (especially in DVB-T; DVB-S upgrade in DVB-S2 8PSK= more bandwidth on same transponder)

There is no point to optimise studio encoders for 4-10 Mbps. These encoders are used with 12 + Mbps and that the bitrate range what they should be optimise for. Broadcaster encoder is a totally different story.

Yes, but you will see when a client want to put lot of bonus on same HD disc
(eg Ghost Rid... european version, studio can not put all HD vide bonus with some codec) ;)
And, generally, client don't want to make more HD disc beause it's more expensive.
12 + Mbps is very good for H264 codec but limit for others if the source is not good.
4Mbps is too low but 6Mbps to 10Mbps for bonus or movie will be interesting.



Lot of people use (used) MPEG2 Digital Vision encoder with DVD because
it can make the best picture display with medium and low bitrate.
I don't see lot of DVD at 8.5 or 9Mbps today on DVD Video disc (because
bonus, DTS tracks, multi 5.1/6.4 audio track,...)

- Or HD Astra demo channel (trailers of differents channel) (if you have a ftp I can upload some TS files at 8.5Mbps).


Even though these trailers were broadcast in a demo channel, they are NOT in the public domain, and are still subject to full copyright protection.
So, please don't offer that contents to me unless you have written permission by Astra and/or the distributor to share that contents.

Even though these trailers were broadcast in a demo channel, they are NOT in the public domain, and are still subject to full copyright protection.
So, please don't offer that contents to me unless you have written permission by Astra and/or the distributor to share that contents.

Yes I know but lot of shop and/other broadcaster professional used/use it.
Some broadcaster/distributor don't right to use or share it (impossible to see all satellite).

And I don't need to upload because this demo is one of demo on P2P ( :( ), but my version work fine
on HD hardware player which support TS (direct record in H264 with 5.1 channel).

But I can give you a H264 demo of ED on TS file with 5.1 AC3 audio stream (I can put a DVB-CPC_ information/flag....).

I want to say that MS will give a "nice" support and will make some encoding...
(Miami Vice was encoded by "nice" people)
It's actually very difficult to do that on a regular basis. Yes, some MS people have helped production houses with encoding certain titles, but this has typically been done to demonstrate features of new PEP versions and it's usually been done on site (at the production house). It's really an exception to the rule and none of this was considered as part of the Paramount deal anyway. The Paramount deal came as a surprise to a lot of us on the Microsoft VC-1 team, actually - we weren't involved in it and didn't know it was coming.

@ Zambelli

When there is an intervention of VC1 team for HDDVD Title,
the video quality of HDDVD Title are everytime the best :D:D
I bought every HDDVD title made by MS team :cool:

If lot of compressionist have the experience of your MS team for VC1 encoding
all HDDVD would have the best technical score on magazine.
MS need to make more technical training for VC1 encoding (but much advanced,
about all video preprocessing options).

Sure if some titles like Shrek,... have the "nice" support of MS team, I do pre-order for this titles today. :p :p :D

Any news from Ben and about the challenge?

I just got back from IBC. I've been on the road the last 3.5 weeks, and haven't had a chance to pay much attention to anything.

Any news from Ben and about the challenge?

Ben promise soon the video on doom9.


I listen (and will see soon, and hope to test) that there is a H264 encoder "Killer" from a H264 expert company.
It can encode with the "best" quality of H264 standard with HDDVD and BD restriction.
It's, with my info, better than all other H264 solutions (for quality). @sspears and Cjplay, the quality is better than x264.
May be this company will make a hardware H264 encoder for HDDVD/BD (because they have already good broadcast encoder).
It will be hard after for VC1.

Hi all,

I'm happy to come back here.

I'm very happy to see in USA a blockbuster in H264 (VC1 wasn't the best codec ?...) with Transfomers.

I surprise to have no news from MS to finish the comparison with their movie source,
on Doom9 we wait since 3 months to say if H264 or VC1 is the best codec for new HD media.

But I sure that lot of studio know what is the best codec (the doom9 challenge showed something ?...).
I hope to see more new titles in H264 and a new H264 encoder will available soon for HD and BD
it is really good (I hope the price will be realist not like Sony, MEI, CC-HDe,...).

Last I hope to see an update of professional tools (HD muxers) for AVC stream because there are some problems
(for VC1 there was one encoder so all muxer were compatible but for H264 there were/are not only one...)



Regards,




Golgot13

...

I surprise to have no news from MS to finish the comparison with their movie source,
on Doom9 we wait since 3 months to say if H264 or VC1 is the best codec for new HD media.



I do not expect Microsoft to meet that challenge.

- Tom

I surprise to have no news from MS to finish the comparison with their movie source, on Doom9 we wait since 3 months to say if H264 or VC1 is the best codec for new HD media.

But I sure that lot of studio know what is the best codec (the doom9 challenge showed something ?...). I hope to see more new titles in H264 and a new H264 encoder will available soon for HD and BD it is really good (I hope the price will be realist not like Sony, MEI, CC-HDe,...).

Last I hope to see an update of professional tools (HD muxers) for AVC stream because there are some problems (for VC1 there was one encoder so all muxer were compatible but for H264 there were/are not only one...)
Sheesh, I'm sorry it's taking so long to get the source file edited together, but you're really overreaching here. All that's really going on is that the recent reorg I'm sure you've all heard about has resulted in a bunch of other things I'm working on getting escalated above this particular project. I'm off at Streaming Media West all this week, but I hope to find some time to get the clip done next week. But don't take my word on that - I'm embarassed it's taken so much longer than I intended, but I can't say there wouldn't be some further delay.

Anyway, even if I provided a clip that showed that VC-1 hands down beat every existing H.264 implementation, would that really answer the question definitively? All we can compare is implementations, and those are always evolving. If VC-1 had a slam dunk, H.264 implementers would focus on addressing the places VC-1 did better, and vise versa if H.264 did better.

It's also amusing to think that Doom9 has much effect on how studios make purchasing decisions :). They have professional compressionists who advise on these decisions, and they're focused on workflow as much as anything.

I realize you're frustrated you're not being provided with PEP for free, but let's try to keep things on an even keel please.

snip......

It's also amusing to think that Doom9 has much effect on how studios make purchasing decisions :). They have professional compressionists who advise on these decisions, and they're focused on workflow as much as anything.

I realize you're frustrated you're not being provided with PEP for free, but let's try to keep things on an even keel please.

I thought the folks at Doom9 were busy cracking BD+ ;)

People get religious about some of this stuff. To Golgot13 Microsoft is the great Satan. We all have our biases. :D

...
Anyway, even if I provided a clip that showed that VC-1 hands down beat every existing H.264 implementation, would that really answer the question definitively? All we can compare is implementations, and those are always evolving. If VC-1 had a slam dunk, H.264 implementers would focus on addressing the places VC-1 did better, and vise versa if H.264 did better.
...

Yes but when we last visited the Doom9 shootout the X264 implementation of the time appeared to be somewhat beating the VC-1 implementation of the time. You could choose to demonstrably win the next round if VC-1 was good enough. And I thought you promised to do that.

Or you could choose to discredit the contest.

- Tom

I'm hardly discrediting the value of the testing, just my ability to get side projects done these days :).

Every day I look with shame at that 420 GB file on that HD, just waiting for a few slow days...

Reat assured, whatever the results, folks will argue about them endlessly about them. Some part of thsearguments will be enlightening.

I'm very happy to see in USA a blockbuster in H264 (VC1 wasn't the best codec ?...) with Transfomers.
Well, I assume you don't know why AVC was picked or you wouldn't say what you just did ;).

Well, I assume you don't know why AVC was picked or you wouldn't say what you just did ;).

Why was it picked?

@2Channel
I never said that "MS is satan". But MS is not clear (problem with European Commission... I live in EU).
iHD is really a nice standard (good work of MS) but I don't like when some person don't tell the truth about codec
(thank you Zambelli to update the MS website about codec but it's strange to update with old news/comparison of 2002).


@Ben
It's easy to find last implementation of PeP (I surprise to find some professional tool on hackers website...)
I can not use (web comparison) or talk about new encoder (because the company, where I work, signed a NDA...)
but some people can understand you can do better with VC1 implementation (Sagitaire showed you
there is some problem with VBV which had a impact on quality...).


@trbarry
I really think (wait video since august to finish the comparison) that there will be never a second round between VC1 and H264.
VC1 is KO and all other challengers are other implementations of H264. Before next CES, we will have a new best H264 encoder
(really best than X264, Sony encoder, MEI,...)


@amirm
Hi, first time to see answer of my post from you. About AVC and Transformer, if it's politic choose
I really think it's normal because you do same (?) but I'm sure there is a part for encoding quality on choose.
Last for you, I surprise that VC1 was not on "2007 MSU Codec comparison" to know if VC1 codec is really good,
I prevented MS (Ben and Zambelli) many month before and nothing (I'm not sure that MS want really
want to show at professional how VC1 is good compare at other codecs)
Link MSU website: http://www.compression.ru/video/

Before next CES, we will have a new best H264 encoder (really best than X264, Sony encoder, MEI,...)
Will that one be also used for broadcasting?

Will that one be also used for broadcasting?

No real time so no broadcast (offline, but the file created can be broadcasted is possible).
It can use for HDDVD and BD (but VoD and DVB/ATSC will be possible).

A little ask about VC1 and broadcast in USA: there is VC1 on ATSC broadcasting (?) (may be test ?).

Why was it picked?

Amirm don't answer... because the H264 quality is really better...

I'm hardly discrediting the value of the testing, just my ability to get side projects done these days :).

Every day I look with shame at that 420 GB file on that HD, just waiting for a few slow days...

Reat assured, whatever the results, folks will argue about them endlessly about them. Some part of thsearguments will be enlightening.

Two weeks after and nothing, I saw that you answer at some post on DVDlist (so you're not too busy).
On Doom9 forum, we wait since more than 3 month.
Enlightening: yes it's sure, H264 is really the best codec on new HD support.;)
Lot of studio (in USA or on world) start to use it, more and more for HD DVD
(Transformer will not be alone for HD DVD) and especially BD (Sony don't like MS codec ?).

May be the HDDVD 3 layers (51GB) will solve the problem with VC1 codec (space/bitrate)...

Thank you Amir and Ben (and the rest of the Microsoft team) for developing VC1.
It is the best codec.

Amirm don't answer... because the H264 quality is really better...
I already said that H.264 was NOT picked because its quality is better. So to the extent you were waiting for me to say something and believe that, then I have already refuted your assumption ;) :). Repeating the same line with no data of your own, simply amounts to empty cheerleading of H.264.

As to your larger quests to prove H.264 is better, well, you are welcome to do so without us. The studios have incredible amount of experience with these codecs now. Some 400 titles and counting are in VC-1. Satisfaction rating is incredibly high with our codec, encoder and support (yes, it is all three that matters). That is why you continue to see movies in VC-1 with perfect review scores like Shrek. Winning an argument with you one way or the other, is immaterial. If you want to declare that you have won some kind of internet bet, be my guest. I don’t think I, Ben or anyone else would have our feelings hurt :). Our work speaks for itself better than any artificial benchmark would ever do….

VC1 appears to be a better codec. Even at upper 20s bitrates AVC seems to have issues with fast motion and explosions. It might just be the tools, but presently VC1 seems to be a better codec.

As to your larger quests to prove H.264 is better, well, you are welcome to do so without us. The studios have incredible amount of experience with these codecs now. Some 400 titles and counting are in VC-1. Satisfaction rating is incredibly high with our codec, encoder and support (yes, it is all three that matters). That is why you continue to see movies in VC-1 with perfect review scores like Shrek. Winning an argument with you one way or the other, is immaterial.


You said about Transformer and H264 it was a politic choice.
I could said that the 400 titles in VC1 it's same: politic and money (your encoder was/are free = no charge).
On first time, I was only to know if VC1 was a good codec and compared it at others
but after with your comments on website (without proof) about VC1/H264/MPEG2
I was only to show the true :p
I thank Zambelli to update your website about VC1 (especially for H264 comparison)
but update a website last month with information of 2002 is surprising...

So MS team don't want to proove that VC1 is the best codec.
But with the withdrawal at the round 2 of Doom9 codec challenge can be interpreated differently:
VC1 fanboy = MS don't need to proove anything
H264 fanboy = MS can not proove that VC1 is better than H264

About VC1 and H264, VC1 is good because it need less CPU power (e.g.: HDi resize function on XBox360
with HD video in H264 it's hard but in VC1 no problem...).


Amirm, I can say only at MS thank you for HDi but for VC1 you know what I think :rolleyes:...
The quality of VC1 title come from preprocessing and HVS filter (remove grain on dark,...)
Some people on Doom9 forum show that MPEG2 is near of VC1, but H264 can not be compared
(only the problem of time encoding).

So H.264 is the best codec then. And what makes it the best? Is there some algorithm that VC-1 lacks? Just curious.

So H.264 is the best codec then. And what makes it the best? Is there some algorithm that VC-1 lacks? Just curious.

Here's my take on this subject, for whatever that's worth.

There are three codec options in HD DVD and BD players. Mpeg2, Mpeg4 (H.264/AVC) and VC-1.

No one is arguing any more that Mpeg2 is the best codec. That hasn't happened since Don Eklund went around at the launch of Blu-Ray saying that Mpeg2 was superior to VC-1 and Mpeg4.

Blu-ray Movies Hit Stores Today
http://pcworld.about.com/news/Jun202006id126163.htm

Sony already had encoding tools in place that could easily be adapted to Blu-ray Disc. "We started working with Sony Japan a couple of years ago developing an MPEG-2 encoder specifically for this format," says Eklund.

However, Eklund says the company has discovered other advantages to using MPEG-2. "Since then, we've also done some testing with VC-1 and MPEG-4 AVC. We're finding they have some advantages when encoding at very low bit rates, but those low bit rates, as compared with MPEG-2, do not yield transparent picture quality to the original master. When you're encoding, you need to encode the noise that is part of the film grain of that master as well."

So now the argument comes down to whether the best codec is Mpeg4 (H.264/AVC) or VC-1. Both sets of encoding tools are still going through continual revisions, and based on which revision of VC1 or AVC tools you're using, one may come out slightly better than the other when using benchmark tools to judge.

From the discs I own, I like the look of the VC-1 titles a little better, but they're both very good. That's my two cents.

You said about Transformer and H264 it was a politic choice.
No, I didn't say this. I said it was NOT picked because it was technically the best codec as you claimed. Politics had nothing to do with their choice. Nor did anyone pay them to use AVC. If this is not enough hint for someone to figure out why, well, then there is no hope for you :).

I could said that the 400 titles in VC1 it's same: politic and money (your encoder was/are free = no charge).
I am sure you are right. I mean we must pay folks to have our own insider post houses say this about the latest revision: http://www.avsforum.com/avs-vb/showpost.php?p=12209276&postcount=2154

One word, LOTS. The statement is 100% valid. I'm using the newest encoder, now officially called CineVision PSE, as I type this.

The biggest improvement definitely is the reduced amount of tweaking needed by the compressionist. Also, there is far greater control over banding. It could have been handled with prior encoders, but the new version makes it much easier to correct it, if needed.

I'm not sure if I can mention my new favorite feature without sharing too much. But WOW, it blew me away and made my job much easier.

So MS team don't want to proove that VC1 is the best codec. But with the withdrawal at the round 2 of Doom9 codec challenge can be interpreated differently:
Let's be clear about one thing: "MS" did not participate in doom9 anything. Ben had chosen on his own time to play with that. If he wants to continue to do so, he can. But nobody from Microsoft is aware or cares about this benchmark (other than me reading your posts here).

Some people on Doom9 forum show that MPEG2 is near of VC1
Well, they are better experts than 99% of the experts in the world then :).


, but H264 can not be compared
(only the problem of time encoding).
Right. The loop filter in H.264 can soften things way too much for it to be comparable to MPEG-2 :D.

Right. The loop filter in H.264 can soften things way too much for it to be comparable to MPEG-2 :D.

In my opinion, the different in-loop filter is one of VC-1's significant advantages over H.264.
davidcw8

IMHO, based mostly on the Doom9 shoot-out, the X264 AVC encoder currently has a slight edge over VC-1. Other AVC encoders may not.

- Tom

IMHO, based mostly on the Doom9 shoot-out, the X264 AVC encoder currently has a slight edge over VC-1. Other AVC encoders may not.

- Tom

How current is the VC-1 encoder Doom9 is testing with?

So H.264 is the best codec then. And what makes it the best? Is there some algorithm that VC-1 lacks? Just curious.

I bet it comes down to how well each codec handles grain. VC-1 seems to handle clean images pretty easy but the noisy stuff can get pretty messy if not fed enough bitrate (ie MI:3, WTC, We Were Soldiers). Filtering the grain is not the answer.

IMHO, based mostly on the Doom9 shoot-out, the X264 AVC encoder currently has a slight edge over VC-1. Other AVC encoders may not.

- Tom
As far as I remember, the current "shootout" was done using synthetic source material (computer generated images, absolutely noise-free). I'd like to see how both encoders perform with natural sources. However, RBFilms "Nature's Journey" tells a lot of how VC1 and AVC encodes look, if the encoding is done with great care: almost identical for the nitpickers, and more importantly: identical for the average HT fan.

As far as I remember, the current "shootout" was done using synthetic source material (computer generated images, absolutely noise-free). I'd like to see how both encoders perform with natural sources. However, RBFilms "Nature's Journey" tells a lot of how VC1 and AVC encodes look, if the encoding is done with great care: almost identical for the nitpickers, and more importantly: identical for the average HT fan.

Actually Natures Journey was made with VC1 for both encodes.

I bet it comes down to how well each codec handles grain. VC-1 seems to handle clean images pretty easy but the noisy stuff can get pretty messy if not fed enough bitrate (ie MI:3, WTC, We Were Soldiers). Filtering the grain is not the answer.

So 300 was a clean source? ;)

Actually Natures Journey was made with VC1 for both encodes.
Ah, okay. I mixed that up. Well, there are other examples, e.g. "Coming to America", "Flags of our Fathers", "The Prestige".

Actually Natures Journey was made with VC1 for both encodes.

Also I think Natures Journey was done from interlaced source so the results would be less relevant for a comparison of two codecs to be used for highdef DVD's.

There has been an ongoing problem for shootouts that NO 24p quality movie (film) quality material has ever been available for shoot-outs since it is all copyrighted. Ben Waggoner supposedly manage to acquire some but it seems doubtful that will be available for comparison now.

Maybe RBFilms can get some. I would personally pay for either an HD DVD or BD containing some raw 4k scan of only a short segment of any quality 24P film movie shootout material. (even if excessively copy protected in all possible fashiions ;) )

- Tom

Well, they are better experts than 99% of the experts in the world then :).



No on Doom9 forum, people compare the codec with measure tool, PSNR and SSIM
(like "professional, not like MS on old website:
"In Microsoft's own internal tests, VC-1 performs 2 to 3 times better than MPEG-2. In other words, to achieve a given PSNR, MPEG-2 requires a bit rate up to 3 times higher than VC-1. These results were measured using both low-motion and high-motion video sequences. Microsoft also compared VC-1 with H.264 and found that both codecs have comparable performance when PSNR is plotted against bit rate.")
But between VC1 and MPEG2 there is only 1dB with PSNR at 12Mbps (result of codec challenge on Doom9),
so it was impossible to say that VC1 is two or three time better than MPEG2 :mad: !!!!!
MS didn't give any information about measure :rolleyes:
But on Doom9, we give all name of software used: encoder and measure tool (VC1 stream come from PEP 1.06).
On your old website you didn't see any information about software used and your result,
only your conclusion: ~ "VC1 is best"....


I proposed at Ben and Zambelli to include VC1 (from PEP) at MSU 2007 comparison.
http://compression.ru/video/codec_comparison/call_for_codecs_07.html
MSU is the Moscow State University, expert of video (a part of notoriety of H264 codec
from Ateme come from this expert: 2005 codec comparison).
I proposed many month ago, may, but now the codec comparison is closed since september....

You can read on the website of MSU a comparison between MS HD Photo and JPEG2000.... :p


Today, I never see a true comparison between VC1 (for HDDVD use) and other codecs and
we can not have access a true comparison between your VC1 codec and others,
so we can not conclude anything between VC1, MPEG2 and H264 codec about quality and efficiency....

But the industry use H264 for broadcasting and no VC1. I interpret this that H264 is best codec
to broadcast HD channel by satellite and terrestrial (HD channels in europe use H264 codec)
so it's better than VC1 for broadcasting (today, may be a PEP 2.x could change it)...


About x264 and movie with grain: today x264 don't give good result with movie with grain.
But some product available like Mainconcept core (used on Cinevision) work well and "kill" MS codec
(and some new version, not available, can do better than Mainconcept or x264 with CG video).

.........................................

....(and some new version, not available, can do better than Mainconcept or x264 with CG video).

Do you mean this "new version, not available" can do even better than new x264+the recent DarkShikari's optimizations ??.

Do you mean this "new version, not available" can do even better than new x264+the recent DarkShikari's optimizations ??.

Yes, last week but I will retest for you this week.

No on Doom9 forum, people compare the codec with measure tool, PSNR and SSIM
You know we are never going to get anywhere if you keep insisting on replacing people's eyes with computers when it comes to evaluation of video quality. Our understanding of human vision system is so primitive as to only be of modest use in such comparisons. This is why we did not rely on PSNR measurements to design VC-1 despite its heavy usage in MPEG/IT world in design of MPEG-2/MPEG-4/AVC. In more than one occasion, we used an algorithm which had lower PSNR benefit but had better visual results.

The above is why DVD Forum and BD made their selections not with PSNR, but with real double blind tests of the codecs, with their own movies and their own experts.

There are scientific measurements for everything in audio and video. But you don't see people on this forum picking a speaker because it has a better measured response. Instead, they go and listen to it because they know the measures are one thing, how a speaker sounds is a completely different matter.

This is why you don't get me excited over your tests. And oh yeh, we are biased wrt to VC-1. We designed the darn thing so have passion around it and are not going to say it is no good. Likewise, you all have fun playing with AVC all day long so want to prove to the rest of the world it is the best codec. So we are even on that front :). Therefore, what is left is the result in front of us: VC-1 nearly matching 24mbit/sec MPEG-2 performance at 7.7 mbit/sec, and AVC not in DVD forum tests. All the titles in use using VC-1, etc. Those are results which are not subject to bias....

You know we are never going to get anywhere if you keep insisting on replacing people's eyes with computers when it comes to evaluation of video quality. Our understanding of human vision system is so primitive as to only be of modest use in such comparisons. This is why we did not rely on PSNR measurements to design VC-1 despite its heavy usage in MPEG/IT world in design of MPEG-2/MPEG-4/AVC. In more than one occasion, we used an algorithm which had lower PSNR benefit but had better visual results.

The above is why DVD Forum and BD made their selections not with PSNR, but with real double blind tests of the codecs, with their own movies and their own experts.

There are scientific measurements for everything in audio and video. But you don't see people on this forum picking a speaker because it has a better measured response. Instead, they go and listen to it because they know the measures are one thing, how a speaker sounds is a completely different matter.

This is why you don't get me excited over your tests. And oh yeh, we are biased wrt to VC-1. We designed the darn thing so have passion around it and are not going to say it is no good. Likewise, you all have fun playing with AVC all day long so want to prove to the rest of the world it is the best codec. So we are even on that front :). Therefore, what is left is the result in front of us: VC-1 nearly matching 24mbit/sec MPEG-2 performance at 7.7 mbit/sec, and AVC not in DVD forum tests. All the titles in use using VC-1, etc. Those are results which are not subject to bias....

One of the nice things about a truly open codec shoot-out would be the public availibility of ALL the test results. That is, we could all see and verify the PSNR, SSIM, and the encoded results of the same material at the same bit rates with multiple codecs. This gets around the problem of distrusting metrics like SSIM.

- Tom

Also I think Natures Journey was done from interlaced source so the results would be less relevant for a comparison of two codecs to be used for highdef DVD's.

There has been an ongoing problem for shootouts that NO 24p quality movie (film) quality material has ever been available for shoot-outs since it is all copyrighted. Ben Waggoner supposedly manage to acquire some but it seems doubtful that will be available for comparison now.

Maybe RBFilms can get some. I would personally pay for either an HD DVD or BD containing some raw 4k scan of only a short segment of any quality 24P film movie shootout material. (even if excessively copy protected in all possible fashiions ;) )

- Tom

Hmmm... thinking about this Tom, RED Digital Cinema just released the beta of REDCine and REDAlert post-processing apps openly. And Jim has released some short 4K footage pieces openly. If nothing else, that gives us the opportunity to get DPX or TIFF sequences with unencumbered 4K source.

Stacey already has previously done some of his own encodes of RED aquired material, but using 1K sources. I've asked him if CineVisionePSE/PEP can ingest a 4k source, but don't know that yet (I suspect not). RedCine should be able to downres...

One of the nice things about a truly open codec shoot-out would be the public availibility of ALL the test results. That is, we could all see and verify the PSNR, SSIM, and the encoded results of the same material at the same bit rates with multiple codecs. This gets around the problem of distrusting metrics like SSIM.

- Tom
But you still don't get anywhere. All BD/HD DVDs are hand tuned and optimized. Above testing will be automatic so it will not have good correlation with what people care about here. And even with automatic testing, the codecs have many parameters. People can go as far as using different settings for each part of each frame! Some people will have time on their hands and can do this. Others will go for a quicker encode with more default parameters.

Then there is the selection of the source. Who says that on the next clip, another result will be generated?

I could go on but net, net, none of these tests are definitive. We for example, handed out at IBC conference a 24p movie clip, together with exact parameters for the AVC encoder and VC-1, showing that we preserved more of the grain and fidelity of the source. Did it settle anything? No. Folks still went on to use to AVC as was the case in DVD Forum where AVC finished last on a number of the tests.

Now, if we are still developing codecs, then these techniques are useful. But when codecs are fully in the marketplace, then we need to go by breath of that experience, not one-off tests which are difficult to pull off without screwing up the results of one versus the other. There are no real points to be won with these tests to merit the high expense of doing them right.

The way we win new VC-1 designs is simple: customers come to us with their movies, and encoding parameters. They give the same to our competitors and then they judge the results, not us. And if we lose, we lose. Because that is how it should be done, not using some synthetic content, wondering how it applies to your work or whether you have an unknown person passing judgment or your own experts....

Well I agree that blindtests is the most interesting test to do, that way you get around the placebo effect.

Interesting that amir said that VC1 at 7,7 nearly matched 24mbit mpeg2.

My own encodings of HDV to WMV has shown that 6,67mbs isnt enough but 9,67mbs is very close.

But you still don't get anywhere. All BD/HD DVDs are hand tuned and optimized. Above testing will be automatic so it will not have good correlation with what people care about here.

I thought the very latest VC1 could do most things automatically? Things that previously had to be hand optimised (ie. by looking at the 400 titles done so far it now does much more stuff automatically).

I think they could specify exactly what parameters were put into both encodes?

Now, if we are still developing codecs, then these techniques are useful. But when codecs are fully in the marketplace...
Is it too late to change the codecs? ie. to add support for greater resolutions, colour palettes, frame rates, etc?

I think you said that the codecs themselves could work with these (or some of these) but that the players couldn't. Is it too late to make any such changes (to allow players/future players to be compatible with these updates to the codecs?)? If we are going to make changes I would think it would be easier now rather than later when there are more players etc. If it was done now maybe it would just require firmware updates or something? Also, if 51GB triple layer discs get approved this may enable us to do more stuff in the codecs than was previously possible.

Also (not to do with codecs but to do with player specs), for people who own dual format players, would the capability of either disc format be practically the same? ie. couldn't a Dual Format machine use the same bandwidth, processing speed, PiP functionality etc. regardless of which format they were playing?

I can't help but wonder whether politics got the best out of the latest
attempt to do the VC1 vs. AVC comparison (talked about on doom9).

Does anybody know a way to get Stacey interested in such shoot-out? :)

I think it would guarantee the best VC1 encoding (at the given time) and
how good quantitave measurements are can be discussed afterwards.

Diogen.

I thought the very latest VC1 could do most things automatically? Things that previously had to be hand optimised (ie. by looking at the 400 titles done so far it now does much more stuff automatically).
Everything? No. But it can do more than it ever could. The job of the compressionist is still safe though :).

I think they could specify exactly what parameters would put into both encodes?
You can't really as the codecs have different designs and knobs. People have tried to fix shared parameters but then one side or the other says that is not as appropriate for their codec.

Is it too late to change the codecs? ie. to add support for greater resolutions, colour palettes, frame rates, etc?
The answer to fist question is yes, it is too late. While a codec like VC-1 runs just fine in software in the PC and hence could be changed there, changing the bitstream would break hardware implementations so it is not allowed. If you change it, you have to call it something else.

The second part is mostly part of current designs. VC-1 for example, doesn't care what the resolution is so you could it feed it 10k by 10k resolution and it would encode it. The HD DVD/BD formats however, put hard contraints on these parameters so once more, if you exceed them, the player can crash.

If we are going to make changes I would think it would be easier now rather than later when there are more players etc.
Resetting the installed base to zero would not be something people want to do, especially when we are talking about features that a handful of us appreciate. We have increased the video resolution by 6X and yet, still don't have mass market. Going beyond this is not likely to do generate new revenue so the business motivation is missing to make machines incompatible.


Also (not to do with codecs but to do with player specs), for people who own dual format players, would the capability of either disc format be practically the same? ie. couldn't a Dual Format machine use the same bandwidth, processing speed, PiP functionality etc. regardless of which format they were playing?
Dual format player will have the maximum rating for both formats. So yes, you could in theory have an HD DVD with the same rate as BD and its video decoder can handle it.

In general, the best hope for better than BD/HD DVD comes in the form of digital distribution to PCs! Here, we can easily roll out new features and with software players, get them into user's hands quickly and without a decade of pushing things through standards group.

In more than one occasion, we used an algorithm which had lower PSNR benefit but had better visual results.

It is true but when the dB between two codec is low (no more than 1dB).


The above is why DVD Forum and BD made their selections not with PSNR, but with real double blind tests of the codecs, with their own movies and their own experts.

Yes but it depend of encoder and preset. I'm sure MS team prepare nicely their video
but for H264 and MPEG2 I doubt (I saw lot of time bad quality on show from good companies
because it was prepared by marketing staff....)


There are scientific measurements for everything in audio and video. But you don't see people on this forum picking a speaker because it has a better measured response. Instead, they go and listen to it because they know the measures are one thing, how a speaker sounds is a completely different matter.

But on your old website:
"In Microsoft's own internal tests, VC-1 performs 2 to 3 times better than MPEG-2. In other words, to achieve a given PSNR, MPEG-2 requires a bit rate up to 3 times higher than VC-1. These results were measured using both low-motion and high-motion video sequences. Microsoft also compared VC-1 with H.264 and found that both codecs have comparable performance when PSNR is plotted against bit rate."

you use a scientific measure measurements (on Doom9 we use same).
About test, "visual" test, MSU use it too to give a better conclusion, but you don't like
that VC1 is on a scientific comparison challenge, aka MSU comparison.



This is why you don't get me excited over your tests. And oh yeh, we are biased wrt to VC-1. We designed the darn thing so have passion around it and are not going to say it is no good. Likewise, you all have fun playing with AVC all day long so want to prove to the rest of the world it is the best codec. So we are even on that front :). Therefore, what is left is the result in front of us: VC-1 nearly matching 24mbit/sec MPEG-2 performance at 7.7 mbit/sec, and AVC not in DVD forum tests. All the titles in use using VC-1, etc. Those are results which are not subject to bias....

Like on your old website, you give some information (codec and bitrate)
but we can see nothing and we can reproduce it....
I really want to see your files (VC at 7.7Mbit/sec and MPEG2 at 24Mbit/sec)
and want to know who made the MPEG2 encoding (may be MS team...).
Next time, if there is some comparison like this, put the file on Doom9 forum
and you will surprise by the other codec that VC1, and you will can really give
a true conclusion...

We have increased the video resolution by 6X and yet, still don't have mass market. Going beyond this is not likely to do generate new revenue so the business motivation is missing to make machines incompatible.
5x more than UK DVDs maybe ;)
Also, concert DVDs on PAL were 50hz. Now we're going to get 24p.:(

In general, the best hope for better than BD/HD DVD comes in the form of digital distribution to PCs! Here, we can easily roll out new features and with software players, get them into user's hands quickly and without a decade of pushing things through standards group.

In general I'd like machine to be able to have even higher resolutions, frame rates, colours palettes, bandwidth, but I'm not sure PCs are currently the best.

For one thing, PCs don't currently have the disc space compared to the 30GB-51GB capability of HD media (I mean a standard hard drive would easily fill up once a few 30-50GB films were stored). It would also take yonks to download at my internet speed. Basically I'd rather have a box that was intended for playing back HD media than it being a PC.

Yes but it depend of encoder and preset. I'm sure MS team prepare nicely their video but for H264 and MPEG2 I doubt (I saw lot of time bad quality on show from good companies because it was prepared by marketing staff....)
There you go. No matter what the benchmark, someone complains that their side was not represented well. Never mind that the DVD Forum test probably cost $1M+ to execute and was repeated twice, with two different sets of parameters and still showed VC-1 to finish first (and no less than three separate companies provided AVC encodes). Someone totally unfamiliar with the test, or who produced the content and what their skill set were, comes out and clearly says it must have been wrong! So what hope is there for the benchmark you are pushing? So why keep wasting time on benchmarks when they are dismissed so easily out of hand? Customers have the real thing and they are the ones who should be deciding. Should you ever own a movie, you are more than welcome to publish in AVC :).

And by the way, in the first round, we only had time to encode each clip once and then submit. We had less than a week to do the work when everything else had two months before we entered the process late! So nothing was “prepared nicely” as you put it….

In general I'd like machine to be able to have even higher resolutions, frame rates, colours palettes, bandwidth, but I'm not sure PCs are currently the best.

For one thing, PCs don't currently have the disc space compared to the 30GB-51GB capability of HD media (I mean a standard hard drive would easily fill up once a few 30-50GB films were stored). It would also take yonks to download at my internet speed. Basically I'd rather have a box that was intended for playing back HD media than it being a PC.
Hardwired devices can always follow PCs. Recall that the first HD optical format was WMV-HD, a PC solution. Sigma Design subsequently built silicon to decode it for embedded devices which led them to design wins for BD players!

My point was that only the PC can do this but rather, if you want this stuff to happen anytime soon, the only path is through digital distribution. We developed WMV-HD in matter of months. In contrast, it took 5 years for HD DVD and close to 8 years for BD to get this far....

Reminds me of the Pepsi Cola vs Coca Cola "taste test" done by Pepsi. "Oh, I think I prefer this one -- it tastes cooler and just plain better than the second one. The second one tastes 'flat'." "Congratulations sir, you've chosen the Pepsi."

Of course should Coca Cola run such a taste test it's the Pepsi that tastes flat. I wonder how that can be?

Reminds me of the Pepsi Cola vs Coca Cola "taste test" done by Pepsi. "Oh, I think I prefer this one -- it tastes cooler and just plain better than the second one. The second one tastes 'flat'." "Congratulations sir, you've chosen the Pepsi."

Of course should Coca Cola run such a taste test it's the Pepsi that tastes flat. I wonder how that can be?
Sorry, which test are you talking about? Doom9 or DVD Forum?

I imagine that it's something like having competing companies produce a purely synthetic flavoring. It doesn't matter what molecular structures and component ratios you produce to be measured against the natural flavor. It only matters which one tastes more like the natural flavor.

It's not only a challenge of mimicing the target flavor chemically, it's knowing how small variations in the chemical composition can change how the flavor converts to a taste.

And by the way, in the first round, we only had time to encode each clip once and then submit. We had less than a week to do the work when everything else had two months before we entered the process late! So nothing was “prepared nicely” as you put it….

You said lot of thing that I can not refute.

But there is some contradiction: you used the same measure tool to compare your codec
at other on your old website and today you said it was not the best tool ???


- You say that VC1 is the best codec, it's so why there is lot of HDDVD in VC1.

- I say that VC1 is not the best codec (H264 is really best), PEP is good encoder (with one of best
video preprocessing) and it was/is free and it's so why there is lot of HDDVD in VC1.
But today authoring studio start to make better HD title (they need to down the bitrate to put
more HD bonus or more lossless audio codec) and start to use H264 codec.
Transformer is in H264 and use the Toshiba encoder, why Toshiba made a H264 encoder if VC1 is best
(a guy PM that Transformer use Toshiba encoder on doom9 forum).


About VC1 and H264 quality: if VC1 is the best codec why all industry don't use it on new HD devices:

There are lot of HD camcorder, APN, set up box,... which use AVC (no VC1 ?)
(NHK start SHD in AVC, on Ceatec I saw professionnal recorder in AVC,...).
The industry push the H264 encoder chipset (no VC1 ?).
All HD channel start to broadcast now in H264 (never see broadcast channel
in VC1) but DVB STB HD in europe are compatible with VC1 (ITU have approuve VC1
so all STB support it...).

I can agree with amir's statement that PSNR by itself is not the best tool to measure video quality, but what Amir doesn't say is that PSNR is one of several valid tools that can be used along with several others to help you design better encoders.

The problem with subjective testing is that it doesn't really help the person who has to design the heuristics for the scene-division and rate control. When faced with specific streams, and relying ONLY on subjective tests by eyeballs, the easiest thing would be to borrow bits from somewhere else to plug the problem areas and keep hoping that noone notices the bits you borrowed from have opened up holes in other areas.

This is assuming an attempt to maintain a target bitrate over a certain amount of time. For one, you can't exceed the peak allowed for the duration of the allowed buffering period in your spec.

In reality, Eyeballs is an interative process that never ends, but worse, for rigorous eyeballs, this has very long leadtime between each formal eyeballs double-blind tests. Leave the eyeballs to the end, not as part of the process.

If I was going up against some other encoder for example, and I was told that there were regions where this other encoder performed better, the first thing I have to understand is what does 'better' mean?

Usually, if it is a certain region at a certain time period, it is important to look at those regions for those frames and do a detailed analysis. And yes, PSNR is just one tool you can use to understand why a particular encode came out as better to your customers. I also use difference maps for visual cues, but I feed them to other mathematical measures other than PSNR. There are many VQ metrics that can be used, I can't go into that as it's a trade secret, if you really want to know, just spend 10 years around video codecs and build your own toolset.

Actually, when it comes down to it, if someone believes that PSNR is the best tool to determine video quality, then that's a pretty dry conversation, as PSNR can actually steer you to what is visually unpleasant to users!.

But if the conversations stays in the : all mathematical tools are useless, and only eyeballs matter, that's just room for obfuscation and you never really get anywhere because all eyeballs are different, and the eyeballs really should be the ones who provide the streams anyway, not the owner of a encoder because each encoder have inherent strengths and weaknesses. So that process becomes open to manipulation.

A lot of encodes today are relying too much on the deblocking/inloop-filter type of trick too much to hide defects in the encode. IMO, these smoothing filters can hide the problems but they do not move us closer to the original pixels. In the end, my money is on being careful with the quantization values. The more you quantize, the less bits you need to encode that residual, but the more loss you take on the accuracy of the macroblocks.

Keep the quantization low for HDM, they don't have to live within the same limits as broadcast HD today, so don't try to make HDM capped like HD-broadcast.

As for pre filtering, I'd rather the studios look carefully at their mastering and get better masters than DNR.

@Neo1965

Agree, Eye needed with subjective view when difference is less than 1dB.
On Doom9, between VC1 from PEP and H264 from x264, the difference is more than 2dB !!!!!!
You can download the video and see the difference (H264 win sure).

You don't need to have a subjective view to appreciate the difference of quality!!!!!


The problem is simple: MS don't want to show what it's the true quality of VC1.
VC1 was never really compare by professional expert (like MSU or MPEG Committee).
I can introduce VC1 on some comparison but I need a agreeement of MS
(all company which have PEP sign a NDA...).
There is some nice comparison expert:
MSU: http://www.compression.ru/video/
MPEG: http://www.chiariglione.org/mpeg/

But I sure that it will never be possible to see a true comparison between MPEG2, VC1 and H264. :(:(:(

I really think like some persons on doom9, Sharktooth:
"They dont provide the (in)famous source... they're not participating the MSU codec comparison too...
My guess is they're trying to take some time to "dodge" a smack..."

I can agree with amir's statement that PSNR by itself is not the best tool to measure video quality, but what Amir doesn't say is that PSNR is one of several valid tools that can be used along with several others to help you design better encoders.
True but the topic is not what is helpful to design better encoders. Rather, folks want to do a shoot out with only metric for who "wins" being PSNR and such. We have participated in many codec shoot outs but none have attempted to just use a measure like this for which codec is better. That is my key objection here.

The analogy I used for speakers perfectly holds here. People designing speakers do use test equipment to measure them. But no two people will "benchmark" the speakers that way. What can be used in design, is not the same thing as the sole metric of what is good.

You said lot of thing that I can not refute.

And yet you continue :rolleyes:

snip.....

About VC1 and H264 quality: if VC1 is the best codec why all industry don't use it on new HD devices:

snip........


Why did Sony Pictures release so many BD titles on Mpeg2? because it was the best technology? Isn't Mpeg2 even more broadly supported in the industry? Doesn't that make it the best technology?

Hmmm... thinking about this Tom, RED Digital Cinema just released the beta of REDCine and REDAlert post-processing apps openly. And Jim has released some short 4K footage pieces openly. If nothing else, that gives us the opportunity to get DPX or TIFF sequences with unencumbered 4K source.

Stacey already has previously done some of his own encodes of RED aquired material, but using 1K sources. I've asked him if CineVisionePSE/PEP can ingest a 4k source, but don't know that yet (I suspect not). RedCine should be able to downres...

I'm looking forward to playing with some RED material. It has worried me a bit that the Bayer pattern they use for capture will require some extra filtering and thus have somewhat less than a 4K effective resolution. OTOH I think much of the advantage of 4K is oversampling anyway so there may still be resolution to burn. Can't wait to find out.

- Tom

True but the topic is not what is helpful to design better encoders. Rather, folks want to do a shoot out with only metric for who "wins" being PSNR and such. We have participated in many codec shoot outs but none have attempted to just use a measure like this for which codec is better. That is my key objection here.

The analogy I used for speakers perfectly holds here. People designing speakers do use test equipment to measure them. But no two people will "benchmark" the speakers that way. What can be used in design, is not the same thing as the sole metric of what is good.

Why not suggest a different metric then? There's lots of very good VQ metrics that are less known, some are even based on finding specific areas that are more important than others. Improving encoders is actually not a bad goal, and having such tools can improve microsoft's future VC-1 encodings (assuming you guys are still doing further R&D there). MSFT could use such a tool.

The speaker analogy is kinda of interesting, about 2 decades ago, canadian national science foundation (or some similar govt research body) had an anechoic chamber that they provided for speaker companies to use. This resulted in a few really really good canadian speakers being designed when I was a student.

And yes, the price is no object enthusiasts will thumb their nose at these data and rely on proper auditions (and well they should as their listening room is not an anechoic chamber). But there's still a very active crowd of bargain conscious audiophiles who do read the mags to actually use this data. So I think the consumers actually do want to look at the measurement data. FR for audio equipment including speakers actually do matter to enough people to make a difference.

This is no different from the way gamers on PCs will pick apart 3D winbench numbers to buy their next $600 graphics card out of some silly 120fps 1600x1200 quake or unreal number. These benchmark numbers differentiate products, they help sell stuff for more money, letting people build stuff that people want instead of everyone trying to drive each other out of business by building more of the cheaper stuff noone wants.

The reality is that even on media-centric PCs video quality of encoders is ripe for a Winbench like treatment, we just haven't found the right consumer grade tool. If such a beast exists, the box makers will be touting those numbers everywhere instead of the downward spiral in price and quality that they're engaged in now.

And, even in CNET.COM, dumbed down reviews of HDTVs still include a geekbox of actual numbers on interesting data --- like color accuracy, temperature just to keep the people who like these numbers interested.

In the absence of actual benchmarks, people are left to sell TVs based on contrast ratio and pixel response time, which doesn't really mean much these days. So we still need a good benchmark.

I'm looking forward to playing with some RED material. It has worried me a bit that the Bayer pattern they use for capture will require some extra filtering and thus have somewhat less than a 4K effective resolution. OTOH I think much of the advantage of 4K is oversampling anyway so there may still be resolution to burn. Can't wait to find out.

- Tom

Indeed.

And maybe I wan't clear enough. This is avaialble NOW. I have a 4K clip. And the tools to manipulate it/export it are avaialble now.

Yes CMOS sensors doesnt have the full res, but with the right processing its "good enough"

The problem with CMOS camcorders is the shutters. They use a rolling shutter wich creates wobbles that make images like lightning harder to shoot.

Why did Sony Pictures release so many BD titles on Mpeg2? because it was the best technology? Isn't Mpeg2 even more broadly supported in the industry? Doesn't that make it the best technology?

According to the reasoning from Sony at the time, the biggest and single most important reason was that it was pushed (from what I was told) was because it would require expensive and extensive hardware/software upgrades to switch to another codec in the beginning of BR adoption in addition to a new learning curve which meant time and money that would keep the content out of the hands of customers. Then you also have the "one encode to serve them all" reasoning since you can in fact use the same equipment and to a lesser extent the same software if you use MPEG2 for HDM. Existing MPEG2 HD resolution encodes could easily be recycled for BR meaning it gets on the shelves that much quicker. Sony also made a LOT of money selling MPEG2 encoding equipment to authoring/encoding firms during the rise of DVD so they didn't want to bite the hand that feeds them so to speak. Things have come a long way since those Sun SPARC workstations were used.

Indeed.

And maybe I wan't clear enough. This is avaialble NOW. I have a 4K clip. And the tools to manipulate it/export it are avaialble now.

Cool. Got links? (assuming that's allowed)

- Tom

Cool. Got links? (assuming that's allowed)

- Tom

Footage:4K REDCode clips (http://www.reduser.net/forum/showthread.php?t=5905)

Toolsets: RED Support Page (http://www.red.com/support/)

IMHO, based mostly on the Doom9 shoot-out, the X264 AVC encoder currently has a slight edge over VC-1. Other AVC encoders may not.

- Tom

Thanks for a frank, independent review of these ongoing results. I trust your opinion/assessment over most of the posters here that have an obvious agenda.

I'm of the opinion that each of the three supported codecs have their own merits, and each shines more brightly in different situations. Is that still valid?

Does the X264 AVC encoder have segment reencoding and ROI?

@Amirm
It's not a problem that a codec "win" but that a codec is more efficience.
Authoring studio have many point to judge the codec after picture quality:
Encoding Speed
Recode pass
Segment recode
Quality
Preset for many support or framerate (eg CC-HDe supported only 23.976 and BD format)
...




@trbarry
"IMHO, based mostly on the Doom9 shoot-out, the X264 AVC encoder currently has a slight edge over VC-1. Other AVC encoders may not."

Slight edge?
With more of 1.5dB, it is really good. On Elephant Dream, x264 can make at 12Mbps better quality than VC1 from PeP at 18Mbps !!!!



@MovieSwede
Yes there is some software encoder which have segment reencoding.

The ROI thing is an interesting concept for home media encoding, but until it can be automated with the proper heuristics to do the scene division by a machine, the need of a human operator makes this a doubtful proposition.

In cases of live transmission, you can't even do that operator intervention thing, so as interesting as it is, it can't be used that often. A better tool would be to write something that automatically identifies ROIs and deal with them in special ways.

The ROI thing is an interesting concept for home media encoding, but until it can be automated with the proper heuristics to do the scene division by a machine, the need of a human operator makes this a doubtful proposition.

In cases of live transmission, you can't even do that operator intervention thing, so as interesting as it is, it can't be used that often. A better tool would be to write something that automatically identifies ROIs and deal with them in special ways.

ROI is something thats best works for handtuning. And for any non real time encoding, you shouldnt be without it.

Footage:4K REDCode clips (http://www.reduser.net/forum/showthread.php?t=5905)

Toolsets: RED Support Page (http://www.red.com/support/)

More RED footage:

http://www.freehillproductions.com/RED/

Andrew

More RED footage:

http://www.freehillproductions.com/RED/

Andrew

Indeed.

Just to be clear, however, I don't believe any of these are original 4K clips acquired directly from the camera.

Although many are indeed hi resolution, they all have already been transocoded in to another lossy codec.

Indeed.

Just to be clear, however, I don't believe any of these are original 4K clips acquired directly from the camera.

Although many are indeed hi resolution, they all have already been transocoded in to another lossy codec.

Crossing The Line is full size HD trailer compressed with Apple ProRes 10bit 4:2:2 codec (similar to Cineform), so the quality is very good.
The problem is that you need MAC with Final Cut Pro 6 to play this file, but I assume someone could convert this file.

Andrew

One thing you can generalise is that there are a great deal of x264 fans at that site, and that they don't get as much info as we do from insiders here, hence a very strong hatred of MS (or M$ as many of them refer to it) exists.

MS played a major role in the development of H.264 as the international committee (ITU/ISO) that developed the H.264 recommendation (i.e., standard) had a MS person as the 'editor' of the standard. They use the term 'editor' where most US organizations would use a different term for the lead person responsible for the development of the standard (e.g., committee or sub-subcommittee chairman). Certainly VC-1 was more of an in-house MS development that was AVC, but MS played a major role in AVC development.

ROI is something thats best works for handtuning. And for any non real time encoding, you shouldnt be without it.

True, but if done manually all the way through, the novelty wears out very quickly. If you ever try to do this with your HDV vacation footage to try to half the bitrate, you might realize the hype is not as practical. Now imagine if an authoring house has to proces dozens of movies every week, there's better use of people's time than to do this by habd.

But, it's much better if you have a pass before hand that seeks these regions and gives you a chance to preview them to accept/reject/modify rather than just blindly draw rectangles all over some screen. That manual editing of ROI has nothing to do with art, unless painting by numbers is considered art.

Some day, if the program is smart enough, a day will come when 99.99% of the time, the program will get the regions correct and never miss a required region, or add a useless region.

---

btw, if I understand the actual use of ROI correctly, and I'm speculating since I have not run these tools myself, all it means is that you want to quantize some areas less than others, in so doing, at some point, you have to borrow bits from other areas or even other frames in order to make those problem regions look better.

When you quantize an area less, it means there are more non-zero terms in the post transform values of residuals or intra values, which means they need more bits to be encoded.

ROI might be fun (for a while) if you really want to squeeze bits, but realize that that comes with a penalty --- often other regions not within the ROI get more quantized, hence more overlap smoothing or more inloop filtering. An example would be a person running with perhaps a background of a building with open windows with people in them, if the person running is put in the ROI, it might just happen that the contents of the open windows end up being quantize more, and become softer. Running analysis tools can also find these problem areas (eg : grid patterns in the sky) if people are interested.

I'm a believer of more automated tools to help the compressionist, including post encode analysis to compare against original to flag potential problem frames. The sky with clouds for example, is where a lot of compression mistakes occur, people don't realize that edges of clouds are very important and should always be in the ROI, but I digress.

My final note is that if one had to sit down with a few hours of video (lets say, ones wedding video) and do this kind of mapping out regions stuff, it stops being fun very quickly, so I imagine that the novelty for those authoring house people would have died down long ago... :D

True, but if done manually all the way through, the novelty wears out very quickly.
What is the old saying? Something as boring as watching paint dry? Yet, that is what professional painters do :). We are not building such tools for consumers. They are for professionals. And just like to a mechanic, a unique wrench to do something others can't do, is very important, and is not judged on novelty of it.

Now imagine if an authoring house has to proces dozens of movies every week, there's better use of people's time than to do this by habd.
Well, most consumers would get bored encoding movies too.

But, it's much better if you have a pass before hand that seeks these regions and gives you a chance to preview them to accept/reject/modify rather than just blindly draw rectangles all over some screen. That manual editing of ROI has nothing to do with art, unless painting by numbers is considered art.
Our VC-1 encoder has a number of modes for auto selecting regions of the screen. For example, with one button click, you can select regions which are getting compressed at certain level and then change their characteristic just as fast. So this is not an either or situation.

ROI comes in handy in those occasions where no machine is able to pick the right region and you just want to draw the darn box and say, "make this look better." 99% of the movie would not need such a treatment. But when you need it, it does the job and removes the furstration of trying to tell the computer what to pick.

Some day, if the program is smart enough, a day will come when 99.99% of the time, the program will get the regions correct and never miss a required region, or add a useless region.
Well, it is even better than this in that we hope some day majority of the movie would not need any tuning let alone picking the regions and having you optimize them. Our recent release with its much improved AI has made great strides in this respect. But for now, we need a skilled operator to get the best out of the codec and what we do is provide them the absolute best tools available.

ROI might be fun (for a while) if you really want to squeeze bits, but realize that that comes with a penalty --- often other regions not within the ROI get more quantized, hence more overlap smoothing or more inloop filtering.
The entire notion of creating DVDs, HD DVDs and BDs is based on the same principal and yet, it works. Yes, you do borrow bits from other parts. But the encoder is just as guilty of putting too many bits in segments, as too few. ROI and our other selection tools lets this model extend to withing an individual frame. That way, if you wanted, you can do "micro" adjustment to VBR encoding in addition to "macro" adjustement that the simpler old fashioned encoders do.

An example would be a person running with perhaps a background of a building with open windows with people in them, if the person running is put in the ROI, it might just happen that the contents of the open windows end up being quantize more, and become softer.
There is no reason for anything else to get "softer." That is where the art of compression comes from. You identify where bits of wasted and apply those to areas that need it. Yes, if you don't know what you are doing and take away bits from things that need it, then there will be artifacts. That is why the skill of the operator is key and with folks having done 100+ movies in VC-1, one does not worry about such things.


'm a believer of more automated tools to help the compressionist, including post encode analysis to compare against original to flag potential problem frames. The sky with clouds for example, is where a lot of compression mistakes occur, people don't realize that edges of clouds are very important and should always be in the ROI, but I digress.
If we can flag things correctly, then we can fix them before the operator sees them :). It would be strange for the codec to know what doesn't look good, but not be able to figure out a way to make them better. Detection is by far the harder problem (than balancing bit budget).

My final note is that if one had to sit down with a few hours of video (lets say, ones wedding video) and do this kind of mapping out regions stuff, it stops being fun very quickly, so I imagine that the novelty for those authoring house people would have died down long ago... :D
Our application is not wedding video. The content has far higher value, which means that you can pay people to do such things and do them well, because you get to amortize the value over large number of consumers, than just a few....

nfm

Our application is not wedding video. The content has far higher value, which means that you can pay people to do such things and do them well, because you get to amortize the value over large number of consumers, than just a few....

Actually there are a couple of fanatical wedding video guys that actually spend to much time with the footage then their budget allows. :rolleyes:

Interesting case study:

Bad Santa (a Buena Vista title) includes the Director's Cut (88') and the Unrated Cut (98'). Instead of using seamless branching to fit both cuts of the film, the BD for Bad Santa includes both versions separately.

What's interesting is that the UC is encoded with VC-1, whereas the DC is encoded with AVC.

From the review at Home Theater Spot (http://www.hometheaterspot.com/fusionbb/showtopic.php?tid/139429/):

Both clocked out at respectably high bitrates. In the places I checked using the PS3s clocking, the VC-1 encode actually appeared to peak a little higher, but the AVC seemed to be better sustained. In other words there appeared broader variance between the peaks and dips with VC-1, where AVC was more uniform.

Now obviously, such comparison is still flawed by the time the compressionist had to work on the project and the effort or skill afforded the operator. I watched the unrated cut first in the BD10, which doesn’t indicate codec (I later discovered the difference in codec using the PS3). After which, I spot checked the DC using the same player. I thought, I noticed a subtle improvement in richness and depth and an ever so slightly stronger sense of clarity in the DC. Detail appeared about the same, but the image seemed a little more dimensional possibly from subdued ringing. EE isn’t serious in either cut, but the DC looked a little less artificially edgy and edge halos were harder to find. This might have resulted from the filtering found in AVC codec or it could be something else. Differences weren’t dramatic, but that’s what initially made me suspect that both films may have been encoded separately instead of using branching. Again at this point I had no reason to even suspect that different codecs were used.

I then checked a couple scenes where I noticed some blocking in the backgrounds in the unrated cut. I didn’t see the same blocking in the AVC encoded DC. Compression artifacting posed rare disturbance to the unrated cut as well, but still, it appeared even less of one in the DC based on the parts I looked at more closely – an observation that is consistent with less deviation in the AVC bitrate. Ultimately, I wasn’t that overwhelmed by the look of the film. Both presentations seemed a bit soft and plaid. Even text wasn’t as crisp as most BDs I’ve reviewed lately. But the video was distinctly/quite clearly, if not dramatically superior to DVD in both resolve and clarity. The DC seemed to improve on the unrated cut by a margin. But a closer, side by side comparison may be in order to rule out deficiencies of memory.

"soft and plaid"?

Is that a new filter? :D

What's interesting is that the UC is encoded with VC-1, whereas the DC is encoded with AVC.
From the review at Home Theater Spot (http://www.hometheaterspot.com/fusionbb/showtopic.php?tid/139429/):

Notice the reviewer did not mention the 11% longer running time for UC version, and the no doubt lower average average bitrate as well?
davidcw8

Interesting case study:

Bad Santa (a Buena Vista title) includes the Director's Cut (88') and the Unrated Cut (98'). Instead of using seamless branching to fit both cuts of the film, the BD for Bad Santa includes both versions separately.

What's interesting is that the UC is encoded with VC-1, whereas the DC is encoded with AVC.

From the review at Home Theater Spot (http://www.hometheaterspot.com/fusionbb/showtopic.php?tid/139429/):
You are seriouly referencing a Chad Varnadore review which talks about h264 vs VC-1? Please tell me that you're joking...

:confused:

You are seriouly referencing a Chad Varnadore review which talks about h264 vs VC-1? Please tell me that you're joking...

:confused:

Yea thats pretty sad.

Which Chad Varnadore? The one on bluray.com or hometheaterspot?

Which Chad Varnadore? The one on bluray.com or hometheaterspot?

They're the same person, and he is deffinitely Pro BD.

They're the same person, and he is deffinitely Pro BD.
A pro BD guy could still judge codecs fairly. My problem with Chad is not that he is a pro BD buy. My problem is that he has a history of unreasonably bashing VC-1. If there was a list of VC-1 haters, Chad would have the top spot.

They're the same person, and he is deffinitely Pro BD.

BD plays VC-1 (and at higher bitrates than HD-DVD). So what's the problem?

A pro BD guy could still judge codecs fairly. My problem with Chad is not that he is a pro BD buy. My problem is that he has a history of unreasonably bashing VC-1. If there was a list of VC-1 haters, Chad would have the top spot.

Ah, okay, I see your point. Forget it then. :)

Ah, okay, I see your point. Forget it then. :)
No probs. I would really like to see a Bad Santa DC vs UC comparison from an unbiased reviewer. Sounds like a(nother) good chance for a h264 vs. VC-1 comparison. Although of course we may never know if the same master was used for DC and UC etc etc. So there'll still be excuses for the loser of the comparison (if there is a clear loser).

The question would be why did they use two different encoders if it was the same quality of the sources?

The question would be why did they use two different encoders if it was the same quality of the sources?
Yeah, the "why" would be interesting - regardless of whether the source had identical quality or not... :)

Our work speaks for itself better than any artificial benchmark would ever do….

Well, that's certainly in keeping with Microsoft's usual responses isn't it?

http://www.theinquirer.net/gb/inquirer/news/2007/12/04/microsoft-slammed-attacking

They're the same person, and he is deffinitely Pro BD.
And generally tows the party line... certainly not known for being pro-MS or pro-VC1.

If the same review came from someone a little less biased, I might give it more credit...

Well, that's certainly in keeping with Microsoft's usual responses isn't it?
You guys are funny :)....

Hey, selectively discrediting unfavorable benchmarks has always been an industry standard practice. Microsoft didn't invent that.

Many decades ago some USA car company showed that when Cadillacs (?) collided with Volkswagons it was somewhat unsafe for the smaller car. So Volkwagon ran tests of smashing Cadillacs with buses, showing it was all sort of relative. ;)

I like the Doom9 tests and hope they continue, with better source material and larger participation.

- Tom

Perhaps we can continue with a civil and scientific discussion of VC-1 and AVC in this thread. (I kept on referring to this tread in the thread I started that got locked.)

http://www.avsforum.com/avs-vb/showthread.php?t=969782

In better new, I actually got a couple of hours in on the 24p clip before I had to run off to Florida (currently stuck in the Cincinnati airport). If no new crises present, I may be done by the end of next week.

source (very difficult without pre-process)
ftp://ftp.ldv.e-technik.tu-muenchen.de/pub/test_sequences/720p/720p50_parkrun_ter.yuv

AVC encoding at 12 Mbps
http://jfl1974.free.fr/upload/Parkrun_Elecard.mp4

VC1 encoding at 12 Mbps
http://jfl1974.free.fr/upload/Parkrun_Elecard.vc1


You can see that there is no doubt: For high compression level VC1 can't really fight with AVC.

But for low compression level it's another history and HDDVD/BD use low compression level. For low compression level MPEG2 will make really good result too.

Amirm say that the most important for encoding is not codec itself but more the pre-process. It's true. With good pre-precess the VC1 encoding will produce very better quality. Anyway you can use exactly the same pre-process for AVC and VC1.

At this time I make HDDVD9 encoding (Matrix 1080p at 7 Mbps with special beta AVC build, DD+ at 640 Kbps Fr & Eng, Subtiltes, Menu, Chapter and Bonus) and AVC is able to make encoding at really high quality level (perfect grain retention) but not VC1.

Can you link the VC1 so we can look at it?

source (very difficult without pre-process)
ftp://ftp.ldv.e-technik.tu-muenchen.de/pub/test_sequences/720p/720p50_parkrun_ter.yuv

AVC encoding at 12 Mbps
http://jfl1974.free.fr/upload/Parkrun_Elecard.mp4

VC1 encoding at 12 Mbps


You can see that there is no doubt: For high compression level VC1 can't really fight with AVC.

But for low compression level it's another history and HDDVD/BD use low compression level. For low compression level MPEG2 with make really good result too.

Amirm say that the most important for encoding is not codec itself but more the pre-process. It's true. With good pre-precess the VC1 encoding will produce very better quality. Anyway you can use exactly the same pre-process for AVC and VC1.

At this time I make HDDVD9 encoding (Matrix 1080p at 7 Mbps with special beta AVC build, DD+ at 640 Kbps Fr & Eng, Subtiltes, Menu, Chapter and Bonus) and AVC is able to make encoding at really high quality level (perfect grain retention) but not VC1.

Zodiaque -

Very cool. What pre-processing did you use? Could you post your script?

- Tom

I am popping in here albeit I swore that I swore off this forum a few weeks ago. AVC sems to use the higher bit rates more effectively than VC-1...while VC-1 does a great job at lower bandwidth rates. I will be running some encode tests fo my own to see and making them avalable to the public.

I believe that Ben from MS has agreed to participate. We welcome his support for our testing...as long as it is sincere and they actually call us back to us this time...:) We have heard this promise before and it never happened. We were told by Amir that MS is to busy with "Big Studios" to support someone like us ... even though a promise was made by MS ... they did not follow through in a timely manner and did not support us in the end.

I do not visit this forum often ... may check back in a month ... so don't take it personally if I do not respond to posts. I really don;t have the time these days. AVS seems has come across on the past as primarily focused on supporting VC-1, HD-DVD, MA, and Toshiba. I myself am open to all possibilities for superior performance and ultra fidelity for the ULTIMATE Home Theater presentation. I prefer to remain open minded about technology and innovation moving forward.

VC-1 has its place in the world as does AVC. However, if it keeps getting shoved down people's throats as the one and only greatest thing ever invented superior to anything and everything mankind has ever known ... rather than letting water seeks its own level so to speak ... it may end up just like HD-DVD.

This has always been the MS powerhouse 800 Pound Gorilla Dictator marketing strategy ... telling people what they what rather than listening to what people want .... like Apple and other great companies have done in the past. Not that I like iPods or the crappy sound quality of iTunes ... not by any means. It is pure junk but it does serve today's lifestyle.

In the end, I respect Apple's "Affinity" marketing approach compared to Microsoft's Push Marketing. At least I feel good about making a choice with Apple. I don't mind being manipulkated ... I just respond better when it is done with some class, style and grace.



BD plays VC-1 (and at higher bitrates than HD-DVD). So what's the problem?

source (very difficult without pre-process)
ftp://ftp.ldv.e-technik.tu-muenchen.de/pub/test_sequences/720p/720p50_parkrun_ter.yuv

AVC encoding at 12 Mbps
http://jfl1974.free.fr/upload/Parkrun_Elecard.mp4

VC1 encoding at 12 Mbps
No VC-1 link.

Also, you really need to share the full details of what encoder and parameters you used - the differences between implementations of each standard are bigger than between the two standards.


Amirm say that the most important for encoding is not codec itself but more the pre-process. It's true. With good pre-precess the VC1 encoding will produce very better quality. Anyway you can use exactly the same pre-process for AVC and VC1.
I think you're inaccurately characterizing what he meant - he was talking about preprocessing in the sense of getting a high quality capture, not messing up the scaling, etcetera. Getting preprocessing right is critical for all encoding irrespective of quality (and getting it wrong is the main cause of really bad looking video). MPEG-1 with correctly processed source can beat both VC-1 and H.264 with source that's badly processed.

At this time I make HDDVD9 encoding (Matrix 1080p at 7 Mbps with special beta AVC build, DD+ at 640 Kbps Fr & Eng, Subtiltes, Menu, Chapter and Bonus) and AVC is able to make encoding at really high quality level (perfect grain retention) but not VC1.
Again, you need to specify what VC-1 encoder you're using, and which paramters you're using for both.

I believe that Ben from MS has agreed to participate. We welcome his support for our testing...as long as it is sincere and they actually call us back to us this time...:) We have heard this promise before and it never happened. We were told by Amir that MS is to busy with "Big Studios" to support someone like us ... even though a promise was made by MS ... they did not follow through in a timely manner and did not support us in the end.
Richard, we had a time window of several months where we had compression resources available for you, and you never got us a tape.

Thanks for your comments, I hope you reconsider and post here more often, your opinions are valued by many people who participate here or read this forum. I was thinking along the lines of you statement below about AVC and VC-1. After reading some threads on Doom9 and some other AVS threads I developed this theory about VC-1. It was created for lower bit rate applications, so it does not scale up to a higher bit rates as well as the latest version of AVC. Therefore it cannot take advantage of the increased bit rate that Blu-ray has to offer. I think your test disc will help prove or disprove this theory.

I am popping in here albeit I swore that I swore off this forum a few weeks ago. AVC sems to use the higher bit rates more effectively than VC-1...while VC-1 does a great job at lower bandwidth rates. I will be running some encode tests fo my own to see and making them avalable to the public.

I believe that Ben from MS has agreed to participate. We welcome his support for our testing...as long as it is sincere and they actually call us back to us this time...:) We have heard this promise before and it never happened. We were told by Amir that MS is to busy with "Big Studios" to support someone like us ... even though a promise was made by MS ... they did not follow through in a timely manner and did not support us in the end.

I do not visit this forum often ... may check back in a month ... so don't take it personally if I do not respond to posts. I really don;t have the time these days. Also, I am spending more time in the Blu-ray Forum now. The group over there seems to be less biased and far more open minded to all possibilities. AVS seems has come across on the past as primarily focused on supporting VC-1, HD-DVD, MA, and Toshiba. I myself am open to all possibilities for superior performance and ultra fidelity for the ULTIMATE Home Theater presentation. I prefer to remain open minded about technology and innovation moving forward.

VC-1 has its place in the world as does AVC. However, if it keeps getting shoved down people's throats as the one and only greatest thing ever invented superior to anything and everything mankind has ever known ... rather than letting water seeks its own level so to speak ... it may end up just like HD-DVD.

This has always been the MS powerhouse 800 Pound Gorilla Dictator marketing strategy ... telling people what they what rather than listening to what people want .... like Apple and other great companies have done in the past. Not that I like iPods or the crappy sound quality of iTunes ... not by any means. It is pure junk but it does serve today's lifestyle.

In the end, I respect Apple's "Affinity" marketing approach compared to Microsoft's Push Marketing. At least I feel good about making a choice with Apple. I don't mind being manipulkated ... I just respond better when it is done with some class, style and grace.

Thanks for your comments, I hope you reconsider and post here more often, your opinions are valued by many people who participate here or read this forum. I was thinking along the lines of you statement below about AVC and VC-1. After reading some threads on Doom9 and some other AVS threads I developed this theory about VC-1. It was created for lower bit rate applications, so it does not scale up to a higher bit rates as well as the latest version of AVC. Therefore it cannot take advantage of the increased bit rate that Blu-ray has to offer. I think your test disc will help prove or disprove this theory.
Actually, it would be the other way around - VC-1 development was much more focused on HD encoding and playback than AVC was. AVC was really focused on videoconferencing and other low-rate videos. I doubt that there was really any testing around BD data rates for it while it was being development. VC-1 was developed and tested across a much broader range of bitrates.

Up to what bit rates was VC-1 developed and tested to? From my understanding AVC has evolved and can perform very well at high bit rates.

Actually, it would be the other way around - VC-1 development was much more focused on HD encoding and playback than AVC was. AVC was really focused on videoconferencing and other low-rate videos. I doubt that there was really any testing around BD data rates for it while it was being development. VC-1 was developed and tested across a much broader range of bitrates.