2006 codec comparison

[vlada]

Hello all and Doom9 especially,

The year 2005 is slowly comming to it's end and I think we can start discussing, which codec would we like to see in next test. Also it would be useful to discuss video sources. These are my suggestions:
x264 and last year's winner ND AVC can't be missing. And the 3rd AVC codec - QuickTime. I'm not sure if it makes sense to test DivX, but we should see XviD there. Aslo MS released a new VC-1 encoder and this codec shouldn't be missing either. I don't know if there was a progress i RealMedia video codec. If yes, it should be included too. And the last one - the only free video codec - Theora.

As for sources - one should be a hard to compress movie, but if possible, it shouldn't be a DVD-rip. You think someone could provide a high quality source not impaired by MPEG compression? The second one should be a DVD-rip, maybe a computer animation (like Shrek, Searching Nemo etc). This type of source was never tested. And the last one should be a classic cartoon.

What are your opinions?

Vlada

[Sirber]

IMO:

codecs: x264, ateme, quicktime avc, vp7, xvid, divx 6, RV10, VC1 (latest version)
sources: same as 2005

[Doom9]

the only free video codec?First of all, free is everything you don't have to pay for, so XviD is free, x264 is free, libav mpeg-4 is free, etc. Second, there are other codes for which you don't have to pay licensing fees, like snow, dirac.

Regarding sources I have tested a considerable number of CG animation sources in the past year and found none to be satisfactory when it comes to pushing codecs to their limit. And if you can get your hands on an uncompressed HD source, be my guest. If not, things are going to be DVD rips for the forseeable future.

Codecs not previously tested will have to undergo a qualification program. Candidates at this point are codecs like libav mpeg-4, snow, dirac, theora, quicktime. It takes a considerable amount of time to put a codec through the full test, and I'm not willing to do that for every codec anymore, only those that are serious contestants.

I've also done some encoding during the year so I have an idea which ones of the candidates are unlikely to make the final cut.

[vlada]

Doom9> XviD and x264 are of course free for home use, but you cannot distribute MPEG-4 compressed content for free. This was my understanding of "free" - free for any use. But you are right with Snow and Dirac, I forgot about them. You think one of these codecs can make it to the finales?

For cartoon source - why don't you use a lower bits per pixel ratio? You can try to lower down bitrate or increase resolution. Why not do this?

As for a HD source, unfortunatelly I have no such movie. But I'm sure somebody around here has a HDV camcorder and could provide a sample. It will be MPEG-2 compressed, but with quite high bitrate.

And maybe a noise DV source would be a good test for modern codecs. Did you think about it?

Thank you,

Vlada

[acidsex]

As for a HD source, unfortunatelly I have no such movie. But I'm sure somebody around here has a HDV camcorder and could provide a sample. It will be MPEG-2 compressed, but with quite high bitrate.


Vlada

HDV is NOT HD. HDV takes up no more room than standard DV-AVI hence being able to use a minidv tape to capture HDV and MiniDV to.

[Doom9]

but you cannot distribute MPEG-4 compressed content for free. not true, the first 50k copies are scott free, so unless you're into commercial use, there's no limit.

And the cartoon source will be swapped out with something suitably hard.. no worries there.
Self-made material is a nogo.. I don't want homevideos, I want professional movies.

[Elias]

not true, the first 50k copies are scott free, so unless you're into commercial use, there's no limit.50k as in 50 000? That pretty much makes MPEG-4 free imho. I mean, there's no way I can distribute a webcam encoded MPEG-4 clip to 50 000 people.

[Caroliano]

First of all, free is everything you don't have to pay for, so XviD is free, x264 is free, libav mpeg-4 is free, etc.
The problem here is that in english de word "free" have two meanings: the free of "free bear" and the free of freedom. I think he was speaking about the second one: free of patent free codec, free for all uses, for you see and change the code, etc.

I'm also expeting one more of yours incredible comparisions. I'ts been a long time from the last, that is already quite outdated, for example in relation to AVC codecs.

[webwonk]

All:

I'd like to suggest an additional test to the suite of variables already considered; playback requirement. Specifically some meaningful measure of CPU, RAM, fps - something which indicates how much horsepower is required for a particular encoded file. I understand that less aggresive encoding parameters for a particular CODEC, usually results in gentler playback requirements, but I'd like to see which CODEC and player results in smoother playback as per the testing's usually encoding standards for the source DVD clips.

Additionally, I'd like to see results of how well the new Flash video CODEC (On2, VP6.x) plays back in the FLV 2 container format as called by SWF 8 or SWF 8.5. Like it or hate it; Flash video will become *very* prevalent and it would be great to see how it holds up to steady favorites like XviD, x264 and Ateme's AVC-MP & HP.

Web

[Sharktooth]

@doom9: what about adding a streaming comparison too?

[IgorC]

Self-made material is a nogo.. I don't want homevideos, I want professional movies.

Maybe it will be usefull 41 DVD's with Great Audio and Video Quality
http://www.amazon.com/gp/richpub/listmania/fullview/11LG698ZRNBSI/102-8294976-1515333?%5Fencoding=UTF8
Maybe some HDTV samples? HD-DVD and Blue ray are coming soon aslo HD-DVD rips :)

[smok3]

maybe i can deliver some uncompressed SD pal sources, but not whole movies, how long should the clips be? (i guess you are only interested in whole movies?)

[stephanV]

not true, the first 50k copies are scott free, so unless you're into commercial use, there's no limit.

That's not exactly true, you can distribute 50,000 decoder/encoders for free.

For distribution of content the license is a bit confusing IMO, but if you follow the rules and don't ask money for your content, you don't have to pay anything... I think...

[Doom9]

That's not exactly true, you can distribute 50,000 decoder/encoders for free.I thought there was also a free clause when it comes to distributing content, not only en/decoders. Either way, it's irrelevant when it comes to the codec comparison, it isn't about finding the codec that costs the least.

(i guess you are only interested in whole movies?)Well, that's the scenario that makes the most sense.. results can vary greatly if you only compress a sample clip, then go for the entire movie.

I have zero interest in streaming at this point.

[Inventive Software]

I've had a plan on the cards for about the last 5 months. The problem is getting a meaty-enough machine to carry out the test. :D There's about 20 codecs with 3 DVD sources. Currently it only measures encoding speed, though decoding speed may also be measured. CPU usage is also on the cards, as is cost and value to the end-user.

[ac-chan123]

You search for HD uncompressed Video sources? Why not try the one from ftp://ftp.ldv.e-technik.tu-muenchen.de/pub/test_sequences/ ? The source was here in a threat about a german article in c't which was writen by a member of this Forum.

[bond]

the only 100% free new codecs are ogg theora and dirac, cause they are not using patented technology and/or the patent holders explicitely dont ask for licenses for using their technology

snow wasnt written for being patent- and/or licensefree, so you cant be sure you dont have to pay licenses to some patentholders when using it, still the chances that someone asks you for licenses for using snow are small

[Valeron]

I don't know if anyone here had ever watched an anime named "Karas"~

If you do, you'll certainly agree it's a "death" source for anime/cg test, even more difficult to compressed than "Ghost In the Shell~Innocence".
Episode 1&2's DVD both have been released in Japan.

Even x264@800 x 448 with 2600kbps, the result can't satisfy me.

[FireArse]

the only 100% free new codecs are ogg theora and dirac, cause they are not using patented technology and/or the patent holders explicitely dont ask for licenses for using their technology

Having mentioned the idea of a new release for this forum's test - one of the lead developer's for Dirac thought it a great idea:

"It would be good to be involved in the comparisons. We do have some improvements, mainly to speed, that we could put in a new release.

When are they going to be, though? If they're very soon then it would be best to go with 0.5.3 as the test cycle for a new release takes quite a time.

cheers

Thomas"

Any idea on when the codec needs to be in for? Can I also second the use of 720p & 1080p material for this year's test - to some, 2006 is the year of HD!

1080p 4:2:0 Test material link (http://www.ldv.ei.tum.de/liquid.php?page=70)

[*.mp4 guy]

@Doom9
Do they broadcast any HDTV material in your area? If so then you could use some of that, I get some pretty Impressive free over the air stuff in my area.

[Doom9]

except for the rare stuff on Astra, there's no HDTV at all. And since there are no PCI-Express DVB cards, I can't even capture that.

one of the lead developer's for Dirac thought it a great idea:I wrote Dirac (the user on this board) a PM last week, it's still unread :(

[CruNcher]

You could use the short H.264 HD verification Clips from the TU Munich for a (mini) HD comparsion that the others posted allready :)
And there are also USB solutions for DVB Captureing, im sure some forum users could also send you a 2h record of the Astra HD channels :)

[zambelli]

And if you can get your hands on an uncompressed HD source, be my guest. If not, things are going to be DVD rips for the forseeable future.
So MPEG-2 compressed SD is fine, but MPEG-2 compressed HD is not? Strange logic.
I second that 720p and 1080p encodes should be included in the tests. If a codec can't handle HD in 2006, I'd call that a pretty major disadvantage. Recording and archiving HDTV is becoming common (among HTPC users, at least) in the U.S. so it's crucial that an objective comparison of HD sources exists.

[FireArse]

except for the rare stuff on Astra, there's no HDTV at all. And since there are no PCI-Express DVB cards, I can't even capture that.

I wrote Dirac (the user on this board) a PM last week, it's still unread :(

Um - not sure why they didn't read the PM - I just went to the sourceforge site and quoted this forum's request.

Dirac Sourceforge Forum (http://sourceforge.net/forum/forum.php?forum_id=353618)

@Doom9 - I can get you the direct email to the BBC team if you like.

F

[Doom9]

I sent an email to the address the dirac team used to sign up here.. I hope that'll work out. Can you PM me the address so I can verify it's the proper one?

@zambelli: Well, if you have access to a non corrupted HD MPEG-2 source of say Matrix3, I'd definitely be interested.. same goes for the other movies that'll be in the comparison. As it so happens, I live in Europe and our HD situation is pitiful.. and due to the lack of HD sources on my part (I'm not interested in test clips, I need the real deal - full length movies), how could I possibly to do a HD test?

[darkavatar1470]

I don't know if anyone here had ever watched an anime named "Karas"~

If you do, you'll certainly agree it's a "death" source for anime/cg test, even more difficult to compressed than "Ghost In the Shell~Innocence".
Episode 1&2's DVD both have been released in Japan.

Even x264@800 x 448 with 2600kbps, the result can't satisfy me.
I agree, everything is in black, can't quite get near "perfect" even at those bitrates. but thats pretty much good enough for most pepole. I remember Doom9 citing he needs really BAD looking stuff like screen shots seen in the movies of previous comparison. ( some where in this thread: )
http://forum.doom9.org/showthread.php?t=88250&page=1&pp=20
So I'd suggest trying KARAS for a 175MB encode...

( actually, I think the new Z-gundam movie is harder.... I'm getting a comp test of 0.398 bit per pixel on the unfiltered source (IVTC'd). the artificial film grain is very disturbing...)

(more OT, why are you resizing to 800*448? you're losing vertical resolution & detail. I'd prefer to use 864*480 with some cropping beforehand to make sure my circles are round....)

[Didée]

I don't fully understand the fuzz about HD sources for the next test. Regarding coding technology, it's basically the very same (apart from the not *that* important possibility of tracing " very long" motion vectors). So the major part is (de-) coding performance only. And for that, you can just stackXY() together some normal SD content - voila, there's your HD pretendal, just for the codec. ;)

[darkavatar1470]

maybe a good reason to do encodes with hoz rez > 720 ......

[Inventive Software]

I may get crucified for this, but in the codec test plan I have, I'm gonna include a VideoCD resolution test as well as the DVD sources. It's been a while since I've seen a modern codec test that's included those kinds of sources.

On the subject of HD sources, this may sound totally OTT, but resize the DVD sources to HD resolutions, losslessly encode them again, brush them up if need be, and then use those lossless sources to encode? If there's not enough HD sources to test, then this may be one of the few options that we have.

[bergi]

@doom9
You can ask in the german forum for a full length mpeg2 capture of Spider-Man or MIB, perhaps sombody has still the original mpegs. I already compressed Spider-Man with XviD to 2 DVDs in full resolution. Is it ok for a 720p test?

[vlada]

Inventive Software> Upsizing a movie is not a solution. There will be no details visible, if you do this. So what do you want to compare then? To do a meaningful comparison you need a real HD source.

[zambelli]

@zambelli: Well, if you have access to a non corrupted HD MPEG-2 source of say Matrix3, I'd definitely be interested.. same goes for the other movies that'll be in the comparison. As it so happens, I live in Europe and our HD situation is pitiful.. and due to the lack of HD sources on my part (I'm not interested in test clips, I need the real deal - full length movies), how could I possibly to do a HD test?
I personally don't have a good HD movie at hand - I will let you know if I find one. How about asking around on the local HDTV / DVB / TiVo (http://forum.doom9.org/forumdisplay.php?f=64) forum or AVSForum's (http://www.avsforum.com/avs-vb/) HTPC or HDTV forums? There are plenty of people in the U.S. with HDTV tuner cards who can capture you a nice, clean 720p or 1080i over-the-air broadcast. Shows like Jay Leno are well known for their image quality.

[zambelli]

I don't fully understand the fuzz about HD sources for the next test. Regarding coding technology, it's basically the very same (apart from the not *that* important possibility of tracing " very long" motion vectors). So the major part is (de-) coding performance only. And for that, you can just stackXY() together some normal SD content - voila, there's your HD pretendal, just for the codec. ;)
Oh, thank god, tell all the companies implementing VC-1 and H.264/AVC to call off their efforts! :)
Seriously though... For most codecs HD content invokes an advanced profile, which could mean an entirely different codepath. It's definitely worth testing.

[Doom9]

For most codecs HD content invokes an advanced profileWhy couldn't you use that for SD sources as well? I do fail to see the difference as well.. HD is relevant insofar that one year from now we should be able to get many up-to-date movies in either 720p or even 1080i format (any i format should've never made it into the specs, what do I have all this 1080p hardware for), and HD will eventually make it to Europe in a couple of decades. Are you implying WME uses a different codepath if your source has a horizontal resolution >= 720?

[FireArse]

All,

Having contacted the BBC team:

"Actually, after having thought a bit more, it is probably a good time to do another release. Hopefully we will release 0.5.4 by the end of the week if you can wait that long.

The new release should speed up both encoding and decoding.

Please bear in mind that Dirac is still alpha and, perhpas, not quite ready for the big time. We are aware that usability is not currently its strength. We have focused on getting the algorithm right and speeding it up, but we do hope to address userability and other issues as well. That said there are ways to use Dirac, and hopefully my colleague "asuraparaju" can point some out in another posting."

Quite happy to see the team's continued flexibility, they are a great bunch!

[Valeron]

I agree, everything is in black, can't quite get near "perfect" even at those bitrates. but thats pretty much good enough for most pepole. I remember Doom9 citing he needs really BAD looking stuff like screen shots seen in the movies of previous comparison. ( some where in this thread: )
http://forum.doom9.org/showthread.php?t=88250&page=1&pp=20
So I'd suggest trying KARAS for a 175MB encode...
omg, any episode of Karas compressed to 175MB will only result in an unexpected quality, it's not a dvdrip anymore.
I'd suggest resize it to a large resolution and double the bitrate.
Actually, I care the details of CG&anime mixture scene with motion blur effect more than dark frames.

( actually, I think the new Z-gundam movie is harder.... I'm getting a comp test of 0.398 bit per pixel on the unfiltered source (IVTC'd). the artificial film grain is very disturbing...)
I'm very interested in this one, can you post you test anywhere?
btw, I think "Ms igloo" OVA is a better choice for CG test. I like the story. :D

(more OT, why are you resizing to 800*448? you're losing vertical resolution & detail. I'd prefer to use 864*480 with some cropping beforehand to make sure my circles are round....)
I don't have the DVD, a friend rip and show me the result.

[zambelli]

Why couldn't you use that for SD sources as well? I do fail to see the difference as well..
Considering how many people trust your codec tests and consider them very relevant, I'm frankly surprised that you'd be willing to dismiss an entire new aspect of encoding based on the assumption of "it's all the same anyway".

Are you implying WME uses a different codepath if your source has a horizontal resolution >= 720?
I think you're missing my point. There is a viable possibility that different codecs actually have different code invoked for HD vs SD. There is also the issue of internal color conversion of Rec.601 vs Rec.709 (http://www.spectsoft.com/wiki/RaveManual/Support/Whitepapers/rec601rec709diff). Codecs are just software and it's strange that you would just assume they are perfectly scalable for all content. The problems could be as obvious as some codecs actually being limited in terms of maximum resolution, or as minute as the aforementioned color conversions. If a particular codec weren't HD friendly, I'd consider that a big minus in the year 2006, wouldn't you?

HD is relevant insofar that one year from now we should be able to get many up-to-date movies in either 720p or even 1080i format (any i format should've never made it into the specs, what do I have all this 1080p hardware for), and HD will eventually make it to Europe in a couple of decades.
HD is reality here and now. All primetime TV in the U.S. is broadcast in HDTV. Anyone with a Motorola 62xx set-top-box (the default digital cable STB these days) can capture MPEG-2 TS streams directly from the STB to their computer. If you decide to archive your HDTV shows, it's unlikely you're going to keep them in MPEG-2 - it's more likely you'll slim them down using a more efficient codec such as WMV9 or Xvid.

If you're waiting for HD-DVD and BD to bring HD to the encoding masses, you'll have to wait longer than a year because there's still the little issue of copy protection, remember? The existing community revolves primarily around people ripping commercial DVDs and encoding them to smaller, more efficient formats. It's unlikely anyone will crack AACS within a month from launch so I think it's more likely HDTV will be the primary source of re-encoded HD content within the first few years.

BTW, only HDTV broadcasts movies in 1080i, using the good old fashioned 2:3 pulldown. Movies encoded for HD-DVD or BD will be 1080p/24, just like on today's NTSC DVDs.

[Doom9]

@FireArse. thanks for the email, I had the same on record and I'm now properly in touch with the right people.

If a particular codec weren't HD friendly, I'd consider that a big minus in the year 2006, wouldn't you?Not if you live in Europe or any other place where HDTV is not the usual form of TV transmission. The most common way of watching TV is still SD and that will not change for the forseeable future. Plus, this site still largely caters to people making DVD backups. Obviously, any codec that doesn't properly scale up will have a shortened shelflife, but that shouldn't concern a DVD Backup codec comparison and that has always be the premise.

I have to be the devil's advocate and ask if you're afraid VC-1 is going to come off worse unless you look at the HD angle as well.. nobody else seems to be concerned with this issue at all (and I'm often speaking directly to people who write the code - if anybody should know it's them) and I know most codecs scale just fine. AVC was obviously created with HD in mind, good ASP codecs scale just fine (there's even HD profiles for DivX). And since we have an enterprising bunch of people on this board, I'd expect to hear sooner or later that any major codec has some major deficiencies when you feed it with HD content. If we look at the thin grey line to the release scene, all those HDTV rips use XviD, and there's generally the > SD resolution rip with 5.1 audio as well, so that's a pretty strong signal that XviD has no issues with HD content.

Either way, this is a hobby, I gave a considerable part of my holiday's for this last year, and there is just no way I can include a handful of HD sources as well. Even if somebody paid for it, at this point I'm booked solid until February at work..

[Didée]

Oh, thank god, tell all the companies implementing VC-1 and H.264/AVC to call off their efforts! :)
Seriously though... For most codecs HD content invokes an advanced profile, which could mean an entirely different codepath. It's definitely worth testing.
Sorry, but it's you who didn't get my point, not the other way round. I don't deny at all that HD is a worthwile thing to test codecs on. However the joke is that you don't necessarily need HD content to test a codec's performance on HD content. :)
The central point of HD is just "much detail in big frames". And this you can very well simulate by other means, e.g. what I denoted above. The only drawback of that method is, as already mentioned, that with a real HD source your motion vectors on average are ~200% as long as compared to the trick with stacking-together SD sources. Still, I suppose that doing a test with such a "simulated HD" source, using high quality SD as input, is at least as valuable (if not even more) as doing a test with a "real" HD source where WMV or h.264 already did compress "the crap out of it". Seriously, I've seen more than one "HD" broadcastings where the *effective* resolution had a strong tendency towards being 240*136 (1920/8 * 1088/8) ...

[zambelli]

Not if you live in Europe or any other place where HDTV is not the usual form of TV transmission. The most common way of watching TV is still SD and that will not change for the forseeable future. Plus, this site still largely caters to people making DVD backups. Obviously, any codec that doesn't properly scale up will have a shortened shelflife, but that shouldn't concern a DVD Backup codec comparison and that has always be the premise.
Even though you're in Europe where HDTV is not common yet, I'm sure you know your audience exceeds the boundaries of the old continent.
I'm not going to twist your arm on this one. I made the suggestion, supplied some arguments why I think it's a good idea, but it's your website and your test.

I have to be the devil's advocate and ask if you're afraid VC-1 is going to come off worse unless you look at the HD angle as well..
Absolutely not. I made the suggestion because I thought it'd improve your test. I want to see an objective comparison as much as the next guy.

If we look at the thin grey line to the release scene, all those HDTV rips use XviD, and there's generally the > SD resolution rip with 5.1 audio as well, so that's a pretty strong signal that XviD has no issues with HD content.
Are you talking about the Xvid encoded HDTV rips like you download from torrent sites? Those are only captured in HD - the actual Xvid files are scaled down to SD. I am yet to see a true Xvid HD rip.

Either way, this is a hobby, I gave a considerable part of my holiday's for this last year, and there is just no way I can include a handful of HD sources as well. Even if somebody paid for it, at this point I'm booked solid until February at work..
Like I said, I'm not going to twist your arm about it. Your hard work is appreciated and I was merely making a suggestion that (IMHO) would enhance the tests even more.

[Elias]

I'd like to see Theora in the 2006 codec comparison. Because the bitstream in Theora has been frozen, and although still in alpha state, all future versions will be backwards compatible, so it would be nice to see how much Theora will improve later on from now. I did some encoding tests in the ffdshow build and was amazed of how good results Theora gives. Ogg Theora is definitely to count on for the open source community because it's completely free and not restricted from patents like MPEG-4 is. I think Theora will be very popular in the future.

[zambelli]

Sorry, but it's you who didn't get my point, not the other way round. I never said that to you, I said it to Doom9.

I don't deny at all that HD is a worthwile thing to test codecs on. However the joke is that you don't necessarily need HD content to test a codec's performance on HD content. :)
But the test is not about performance, it's about image quality. :) There are a number of things a codec could do to treat HD content differently than SD content, such as filtering, adaptive block size transforms, YUV-RGB color conversion, etc. If you want reliable results, you'd need true HD content.

[Doom9]

I'm sure you know your audience exceeds the boundaries of the old continent.It does but > 50% lives where HD is merely a dream. Except for the US, Japan and Australia, HDTV is largely a non issue at this time. And I wouldn't be so sure that a majority of people in the US have HD at this point..

Those are only captured in HD - the actual Xvid files are scaled down to SD. I am yet to see a true Xvid HD rip.I've never used bittorrent, but what I'm talking about here is above HD resolution. There's your bread and butter 640x rip, but there are those that approach true 720p. Then again, it's not my scene but I know for a fact that there are such releases that go beyond SD resolutions. I think the resolution was around 1000x .. you can't argue that's not HD, and there are plenty of people on this board using XviD to archive their HDTV rips in HD resolutions.

[IgorC]

I also agree that Xvid is a good codec for HD.
Look at it http://www.megaupload.com/?d=I54RTQY7 it´s 720p 5 mbit/s.
And I´m sure WMV9 can´t do that. Maybe VC-1 will be better but I don´t think that much. If one codec is a good on SD it will be good on HD too.
There was quite good metrics test http://forum.doom9.org/showthread.php?t=90784

So I would suppose there won´t be big changes since then.

TOP10 (for 1 cd rip bitrate)

1. Ateme H.264 HP
2. x264 HP ( can be on par with Ateme H.264)
3. VP7.0.8 . Maybe there will be new wersion?
4. VP6 / ( future Xvid 1.2 cvs ?)
5. Xvid 1.1 / Divx6
6. RV10 / Libavcodec MPEG-4
7. WM9. New VC-1?
.....

But I´m sure Doom9´s new test will be more than interesting. :)

[thana]

I've never used bittorrent, but what I'm talking about here is above HD resolution. There's your bread and butter 640x rip, but there are those that approach true 720p. Then again, it's not my scene but I know for a fact that there are such releases that go beyond SD resolutions. I think the resolution was around 1000x they are called HR-HDTV (http://en.wikipedia.org/wiki/Half_Resolution_High_Definition) and are normally resized to 960x528. the one's which are called HDTV are resized to something lower (usually between 640 and 608). both are captured from hdtv-source.

[SolarBurn]

I could probably provide some uncompressed video from a digibeta source via component video if you want. Wouldn't be movie footage but it'd be professionally recorded. Not sure how much value that would have but it's probably the cleanest SD source you'll get outside an sdi transfer.

[MediaRay]

To IgorC
My Codec's Rank:
1.x264
2.Nero AVC
3.VP7
4.Apple QT7(Although use H.264 codec,but doesn't support H.264 MP)
5.VP6
6.WMV9
7.XviD
8.Real 10
9.DivX 6
......

BTW when Doom9 release the 2005 codec shoot-out?

[IgorC]

In previous comparison there were 2 version of screenshots : png and jpg. The difference between codecs was so evident on png and jpg screenshots.

there is a very tiny difference between png and jpg for schreenshots of videocodecs.

[Doom9]

you are saying? jpeg at 90% quality doesn't look so bad and I'm always using the same program to convert the bmp screenshots. And in my book, QT7 rates all the way down in that list.. you've seen it.. usability nightmare, slow and subpar quality.

[IgorC]

Sorry maybe it wasn't clear from my side. Imo there is no visual difference between jpg and png.

I just wanted to notice. using jpg or png it's the same result because diference between codecs screenshots is more than evident.

[vlada]

Doom9> Thank you very much for the new shoot-out. But I have a question though. Why didn't you use VBR sound for the movies? It really makes sense here IMO. I think it would give some more points to MP4 codecs because of overhead.

[stephanV]

Personally, I don't get the whole dealing with audio and containers. For me it is too irrelevant for a test that is apparently and should be about comparing video codecs. I know this is for achieving a "realistic" back-up scenario, but forgive me, what in earth's name is that? Is targeting 1 CD still the most realistic scenario these days? Is putting DivX in mp4 a realistic scenario? Is using 2ch LC-AAC a realistic scenario? Sure they are all realistic in the sense that it is possible to do this, but is it realistic in a way that these are commonly used methods? That would at least seem highly doubtful to me.

Just let all codecs target the same bitrate and let people themselves decide how exactly they want to back up video. That would confess of much more realism. Muddling with audio and containers only substracts value of the test. Or perhaps it should be renamed to "back-up comparison"...

[Doom9]

Let me just cut you short: everybody who wants to bitch about containers and methodology can go somplace else or faces rule 16. Make your bed and sleep in it.
And for all the know it betters, it is impossible to work with the same container everywhere.. there's always a reason why you can't put codec X into container Y or why you can't work with raw streams.

[Doom9]

here's your one chance to make your voice heard: http://forum.doom9.org/showthread.php?t=104245

[Xayd]

I sent an email to the address the dirac team used to sign up here.. I hope that'll work out. Can you PM me the address so I can verify it's the proper one?

@zambelli: Well, if you have access to a non corrupted HD MPEG-2 source of say Matrix3, I'd definitely be interested.. same goes for the other movies that'll be in the comparison. As it so happens, I live in Europe and our HD situation is pitiful.. and due to the lack of HD sources on my part (I'm not interested in test clips, I need the real deal - full length movies), how could I possibly to do a HD test?

the best HDTV source for such a comparison would be the Gladiator broadcast that was on BEV satellite in canada. it's reference material quality. not even any blurring or pixelation in fight scenes, for all intents and purposes, perfect ;).

and it has good scenes for codec comparison, imo. there's the panning scene at the beginning with flaming arrows being shot into that forest, and another panning scene when they first get to rome and are riding up to the coliseum. also the combat scene with the tigers in the arena floor.

most HDTV these days suffer from down-rezzing and rate-shaping from the cable and satellite broadcasters, so quality is very hit and miss. this broadcast is one of the few flawless ones out there. it's an older broadcast that was done at near-full bitrate before down-rezzing and bitrate shaping became widespread.

if you happen to have access to a premium newsgroup service, it looks like there was a re-mux of this recording put up with a french soundtrack about 20 days ago. newzbin search (http://www.newzbin.com/search/query/p/?q=gladiator+hdtv&Category=-1&searchFP=p)

it's about ~25 gigs for the whole movie.

as for xvid in such comparisons, i think it would actually be surprising to most people how well xvid performs. i encoded the above mentioned movie resized to 720p with a bitrate considered by most to be too low for preserving HD quality (below 4000 kbps, to fit on a DVD-R), and it still looks flawless. the quality lost when encoding to xvid is actually less than the quality lost by software player (libmpeg2 without post processing) resizing to 720p on the fly during playback.

imo, with high quality HD sources, you get to a point where real-time resizing is actually WORSE than something coming out of avisynth that is "properly" resized and re-encoded. look at the below screen captures from this source...

http://img454.imageshack.us/img454/1886/gladiator7zz.th.png (http://img454.imageshack.us/my.php?image=gladiator7zz.png)

one is my xvid encode, the other is the HDTV source file simply played back at a 1280x720 resolution. you would think that the slightly blurred image on the bottom is my xvid, right? it's actually the other way around, the frame on the bottom is the transport stream, the sharper frame on the top is my 4000 kbps xvid encoded with the sharp-mpeg matrix that autogk uses. why is the result what it is? that's arguable, but the result is right there in black and white, it is what it is.

besides the space-saving, i think it's actually feasible that people might re-encode their HD captures anyway, simply because playback software (and hardware) is unreliable. real-time resizing is poor, and software deinterlacing is also poor, on average. so deinterlacing in avisynth along with resizing to the exact pixel for pixel resolution of your display could actually be preferable for quality because playback software/hardware causes more quality loss than re-encoding does.

so yes for codec comparison resolution difference is just a matter of math, but when you get into playback issues that many displays and playback devices/software have, there's more to consider than the simple math of storage space and quality at a certain bitrate. for these reasons HDTV comparison is definitely worthwhile, imo.

[siddharthagandhi]

I suggest the Sony AVC Codec and the Mainconcept AVC codec...both of which found in Sony Vegas.

[baer999]

When will the new codec comparison start or will be released?

[karasu]

good question

[Snakekick]

but no answer

[mdoubledragon]

Yawn! Yawn! So much for the shoot out:rolleyes: