Sample HD MPEG-1 Clip...

[bond]

The Ice Age sample chosen by Sagittaire was encode by him with Ateme ($9K) and by me with MegaPEG.thx, where can i download both clips and the source?

[digigami]

If the source is flawed the encoder cant to miracles.

No argument there. In this case, the Apple source exhibits light banding, so I think a faithful representation of the banding would be what we want.

[IgorC]

http://www.apple.com/trailers/

[digigami]

thx, where can i download both clips and the source?

The source is on the Apple Trailers site. There are two Ice Age trailers so download them both.

Regarding the compressed clips, you'll need to read the thread. Sagittaire's clips are up near the top, mine is around 2/3.

Speaking of clips, I have a lot of stuff to download, and the one web-based download services with the 60s delay is a bit of a hassle.

[digigami]

H264 MP is very good too ... try and see yourself.

Will do.

BTW, all your clips for your MPEG-2 tests are 401 - file not found.

[digigami]

Ateme HP encoder will be enable in Nero7 (approx. less than 100$ ) package in 2006 , while your MPEG-2 codec cost 300$ . I would prefer 0$ mencoder another excelent MPEG-2 with the same or better quality.

That would be up to you. Our encoder has other features for $300.

Referring to products that ship next year is called "vaporware".

[IgorC]

Referring to products that ship next year is called "vaporware".

There is already quite stable Ateme encoder HP version. Some betatesters (most of guys here ;) ) have already tested. It will be here soon :)
there is also x264 opensource with a great quality.

[CruNcher]

@digigami
you could use a raw to avi converter or avisynths raw source plugin

here is the complete verification report
http://www.chiariglione.org/mpeg/working_documents/mpeg-04/avc/avc_vt.zip

as you can see @ that time 2003 HD superiority over Mpeg-2 was by arround the factor of 1.7x today in 2005 we reached the factor of 2x and 3x and it's still improved by different companies as Ateme, Mainconcept/Elecard, Videosoft and many others and not to forget OSS projects like X264 with still projects to come.

[IgorC]

Our encoder has other features for $300.


Like what? Mencoder MPEG-2 is on par with your encoder and it's opensource.

Other features are enable in other open source projects. Like audio mp2, tool to work with MPEG-2 streams and etc.

[savage747]

Since I am in the US, perhaps you can give a breakdown of what HD formats are available via DVB in your area.

HD television just launched regular operation here in Germany. There have been some experimental broadcasts using DVB-S using MPEG2 - but they now are discontinued. All HD broadcasts (I am aware of) now use DVB-S2 (more efficient version of DVB-S - squeeze more bandwith out of one satellite transponder) and h264 - oh, and "nice" encryption.

Basically those programs have zero audience because fitting settop boxes are not available yet (shall be in Q1 2006).

Well, at least they start with the most modern technology... no MPEG2 HD in mid-europe.

<off-topic>
Oh, I took the liberty of encoding some clips with libavcodec single pass (-lavcopts vcodec=mpeg2video:aspect=16/9:vbitrate=2000:vhq) and this already gives much better quality then what can be seen on most DVB-T broadcasts... I wonder what hellspawn of an encoder they use.
</off-topic>

[digigami]

yes. QT7 h.264 isn't ideal source but there aren't so obvious artefacts like on frame 560 where appears blue sky. It's most annoying artefactas I've seen on mpeg-2 video and there aren't on Qt7 1080p source.

So now my question is what video card and if you know, whether VLC is using the video card for IDCT or using the built-in routines. I say this because when I look at VLC in 082 and 084a - not a problem. 24" viewing distance, approx 15" window on 17" display LCD screen. Mac gamma.

On PC, 24" from screen. 17" LCD.

Here are my audition results on Nero Showtime. 17" laptop, full screen playback.

24" H.264 is sharper. Both H.264 and MPEG-2 noticeable banding/blocking and blocking in the blue background behind fight scene and at end when elephant's ass is walking away.

36" H.264 clarity not so evident. Artifacts in cave (dark) not visible on either clip. Blocking/banding in blue sky/blue cave visible in both clips.

[digigami]

and not to forget OSS projects like X264 with still projects to come.

And not to forget...Digigami! ;)

BTW, all you H.264 proponents - you can now consider me to be H.264 proponent. Although I am sure you will not mind if we continue to improve our MPEG-2 encoder.

[Sharktooth]

Good for you. But even better for the designers of the codec.

My request for a lossless version of Ice Age H.264 AVC is to establish the true bitrate that will be used to encode HD-DVD and Blu-Ray. I do not believe anyone will professionally encode for those formats at 1250.

Is there anyone here who can answer this using Ateme? How about the Ateme guy...
no-one will encode professionally HD stuff with your or any other mpeg-2 at 3.3mbps or even at 8mbps as well...
however h.264 can be LOSSLESS too (that's different from transparent).
a professional h.264 encode with all the quality needed for distribution will be something around 3000kbps but you will not notice ANY kind of artifacts even with a frame by frame comparison unlike some actual DVDs.

[digigami]

<off-topic>
Oh, I took the liberty of encoding some clips with libavcodec single pass (-lavcopts vcodec=mpeg2video:aspect=16/9:vbitrate=2000:vhq) and this already gives much better quality then what can be seen on most DVB-T broadcasts... I wonder what hellspawn of an encoder they use.
</off-topic>

Yes, OK, now do it in real-time with capture. Most real-time encoders use 1/2 width frame for motion compensation and do exhaustive search. Also, make sure you are using MPEG standard quant matrices, because I think it is rare to find broadcast with anything else. DVD is another matter.

[digigami]

mpeg-2 at 3.3mbps or even at 8mbps as well...

You are right about that. I recommend to customers to use the highest bitrate they can possibly use. Blu-Ray as a 54 mbits/s ceiling and a big drive capacity. Looking forward to that...

HD-DVD looks like around 27 mbits/s ceiling and 5x the size of DVD5.

[IgorC]

So now my question is what video card and if you know, whether VLC is using the video card for IDCT or using the built-in routines.

What? Now you are saying that´s matter of harware. mpeg-2 has artefactas on my system due to my old Geforce 64 ddr videocard while H.264 is looking better. Is it your point?

Artefacts has nothing to do with hardware. QT7 source video has not such artefacts.

[digigami]

Thanks for all your posts. I'm going to play around with Nero MP and x264.

Like I said earlier, I don't mind being bested by H.264 - and I am glad to have learned that Ateme is a professional solution for the PC.

I will be surprised if they put a $9000 encoder into $100 program at this point in the adoption curve. I guess they worry about x264!

[Sharktooth]

You are right about that. I recommend to customers to use the highest bitrate they can possibly use. Blu-Ray as a 54 mbits/s ceiling and a big drive capacity. Looking forward to that...

HD-DVD looks like around 27 mbits/s ceiling and 5x the size of DVD5.
i corrected my sentence.. i didnt mean "your"... i meant mpeg-2 in general.
Thanks for all your posts. I'm going to play around with Nero MP and x264.

Like I said earlier, I don't mind being bested by H.264 - and I am glad to have learned that Ateme is a professional solution for the PC.

I will be surprised if they put a $9000 encoder into $100 program at this point in the adoption curve. I guess they worry about x264!
They already did it with the main profile encoder.
however they dont provide customer service/care for nero customers... it's up to nero... and nero customer service is aimed at non-professional audience.

[digigami]

What? Now you are saying that´s matter of harware. mpeg-2 has artefactas on my system due to my old Geforce 64 ddr videocard while H.264 is looking better. Is it your point?

Artefacts has nothing to do with hardware. QT7 source video has not such artefacts.

Do you understand that different video cards can handle different amounts of the MPEG decoding pipeline? On the Mac, with the DVD framework, almost the entire decode is done by the hardware. VLC does not access this AFAIK.
QuickTime MPEG-1 does not use this. Thus, on my PowerBook, each of these 3 MPEG decoders gives different results. My PC laptop has high end video card, probably does the entire MPEG decoding in hardware.

http://www.ati.com/products/dtv/revolutionary.html
google search MPEG decoding in hardware ATI video cards (http://www.google.com/search?q=hardware+mpeg+decoding+site:ati.com&ie=UTF-8&oe=UTF-8)

Do you understand?

[Sagittaire]

I will be surprised if they put a $9000 encoder into $100 program at this point in the adoption curve

Ateme encoder is professionnal encoder (interlaced support with transport stream output for example) certainely for industry with small potential client numbers.

Nero suite are not for the same objective ...

[digigami]

What? Now you are saying that´s matter of harware. mpeg-2 has artefactas on my system due to my old Geforce 64 ddr videocard while H.264 is looking better. Is it your point?

Artefacts has nothing to do with hardware. QT7 source video has not such artefacts.

Do you see similar artifacts in H.264? Blue background during fight scene and at the end? If not, you are biased. I can see them in both clips.

[digigami]

Ateme encoder is professionnal encoder (interlaced support) certainely for industry with small potential client numbers.

Nero suite are not for the same objective ...

So the HP in Nero will still not represent the full AVC capabilities? Progressive encoding only? What about other features. Do you know?

[IgorC]

I will be surprised if they put a $9000 encoder into $100 program at this point in the adoption curve. I guess they worry about x264!

I will be surpisied if you put a 300$ encoder into 5$ program. You would worry about mencoder.

[IgorC]

digigami. Nobody will use MPEG-2 to archieve HDs at such low bitrates 3.6-5 mbps. You can make a good marketing and propaganda here but quality isn't here.
You are only the one man here talking about MPEG-2 HD at 3-5 mbps, don´t use. We don't see another guy.

[digigami]

I will be surpisied if you put a 300$ encoder into 5$ program. You would worry about mencoder.

Do you see similar artifacts in H.264? Blue background during fight scene and at the end? If not, you are biased. I can see them in both clips.

Do you?

[Sharktooth]

So the HP in Nero will still not represent the full AVC capabilities? Progressive encoding only? What about other features. Do you know?
Yes it will but i suppose other "pro" features (TS output, interlaced encoding and other stuff meant for broadcast or professional use) wont be there.

[digigami]

digigami. Nobody will use MPEG-2 to archieve their HDs at such low bitrates 3.6-5 mbps. You can make a good marketing and propaganda here but quality isn't here.
You are only the one man here talking about MPEG-2 HD at 3-5 mbps, don´t use. We don't see another guy.

Do you see similar artifacts in H.264? Blue background during fight scene and at the end? If not, you are biased. I can see them in both clips.

[digigami]

Yes it will but i suppose other "pro" features (TS output, interlaced encoding and other stuff meant for broadcast or professional use) wont be there.

Interesting. Thank-you. I plan to try your quant matrices with our MPEG-2 just for kicks. 8x8 DCT is 8x8 DCT.

[Sharktooth]

Do you see similar artifacts in H.264? Blue background during fight scene and at the end? If not, you are biased. I can see them in both clips.
Yes artifacts are there, but the h.264 clip still looks better.
My matrices are for xvid (mpeg4 ASP)... i dont guarantee good results for mpeg2.
Time to go now :)
l8r.

[IgorC]

QT7 source artefactas are nothing comparing to your MPEG-2 sample.
looks like you're nerving right now ;)

[digigami]

Yes artifacts are there, but the h.264 clip still looks better.

Thanks, but I am asking IgorC. He's not evaluating both clips, only mine.

And I agree, the H.264 looks sharper.

[digigami]

QT7 source artefactas are nothing comparing to your MPEG-2 sample.
looks like you're nerving right now ;)

I'm not talking about QT source. I'm talking about Sagittaire's H.264 AVC at 1250. Get a grip.

[JohnV]

So the HP in Nero will still not represent the full AVC capabilities? Progressive encoding only? What about other features. Do you know?
Even I can't say yet. But it should be the same codec with practically all features, interlacing included. Btw. IIRC interlacing is Main Profile feature anyway, not HP.

[digigami]

Yes artifacts are there, but the h.264 clip still looks better.
My matrices are for xvid (mpeg4 ASP)... i dont guarantee good results for mpeg2.
Time to go now :)
l8r.

OK - I will be investigating XviD too for MoviesForMyPod.

[CruNcher]

ok just on the matter of Hollywood and H.264 i think you remember the time when Blue-Ray said hey AVC Main Profile doesn't looks so good as our 20 Mbit Mpeg-2 Stuff

here is the official background to that matter and when Frext was introduced also it should be covered that Atemes Encoder i think as the only one in the world supports Film Grain Modeling in the Encoder/Decoder as a bonus to High Profile.

http://ftp3.itu.ch/av-arch/jvt-site/2004_03_Munich/JVT-K025r1.doc

[digigami]

Even I can't say yet. But it should be the same codec with practically all features, interlacing included. Btw. IIRC interlacing is Main Profile feature anyway, not HP.

Thanks JohnV, I am now bonafide Nero customer for NeroDigital, just waiting for key license key. Can you tell them to hurry? ;)

[IgorC]

Talking about Ateme H.264 , yes I agree with. but are you comparing Ateme H.264 video that´s 3 times smaller than your MPEG-2 video and you´re saying that H.264 is sharper than MPEG-2. Imagine what conlcusion you will do at the same bitrate. H.264 will be much better than MPEg-2.

[digigami]

ok just on the matter of Hollywood and H.264 i think you remember the time when Blue-Ray said hey AVC Main Profile doesn't looks so good as our 20 Mbit Mpeg-2 Stuff

here is the official background to that matter and when Frext was introduced also it should be covered that Atemes Encoder i think as the only one in the world supports Film Grain Modeling in the Encoder/Decoder as a bonus to High Profile.

http://ftp3.itu.ch/av-arch/jvt-site/2004_03_Munich/JVT-K025r1.doc

Thanks. I guess I am Ateme user now that I have Nero. Will refer back to this thread often - at least for the tech info. ;)

[digigami]

H.264 will be much better than MPEg-2.

I guess you haven't been reading the thread, but we all agreed on that about 3 pages ago.

[IgorC]

I'm really happy now :)

[savage747]

Yes, OK, now do it in real-time with capture. Most real-time encoders use 1/2 width frame for motion compensation and do exhaustive search. Also, make sure you are using MPEG standard quant matrices, because I think it is rare to find broadcast with anything else. DVD is another matter.

Well, my encoding speed was something like 64fps (Athlon64 3000+) - so this gives plenty of headroom for capturing. I didn't use a custom quantization matrix (most likely http://www1.mplayerhq.hu/cgi-bin/cvsweb.cgi/ffmpeg/libavcodec/mpeg12data.h?rev=1.28&content-type=text/x-cvsweb-markup&cvsroot=FFMpeg was used - is that standard?). In my experience those files can be played with everything that claims to be able to decode MPEG2... I couldn't find the exact spec (GOP length etc.) for DVB by a simple google query, though (they should be over at ETSI, though - but their site is a mess). I did another quick test (forcing keyframe interval to 10 - which most likely is more restrictive than needed) and the result still looks way better than what can be seen on DVB-T... so I think that much could be done to improve picture quality if the broadcasters would care.

Meh, think I'm fully off-topic now... (SD TV over DVB)

[Sagittaire]

And finally (for me) here little comparison

your best possible encoding (I imagine) at 3800 Kbps
http://digigami.in-long-beach-ca.com/megapeg/Doom9/ice_age_2-tlrD_720p.010.m1v

and my best possible encoding at 3800 Kbps for libavcodec MPEG2 encoder
http://multimediacom.free.fr/Download/MPEG2-HQ5.m2v

Sorry ... but as you can see there are no possible discution ... ;)
Before this thread I thought that Libavcodec was the best MPEG1/2 encoder in the world.
After this thread I think always that Libavcodec is the best MPEG1/2 encoder in the world.
After this thread I think always that H264 is very better than all MPEG1/2 encoder in the world.

PS: Setting for Libavcodec:
- GOP PAL compliant 15 frames max
- 2 Bframes - No adaptative with low ratio
- 128 for Max MV range
- VBV compliant with DVD format
- 20 Mbps for Max Bitrate
- Highest possible ME search and RDO search
- Masking HVS optimisation
- Multipasse Encoding
- Sagittaire Custom Matrix

[digigami]

And finally (for me) here little comparison
and my best possible encoding at 3800 Kbps for libavcodec MPEG2 encoder http://multimediacom.free.fr/Download/MPEG2-HQ5.m2v


Sounds great. Taking a look at it now.

First try with Nero ReCode:
12:09:24 File burn_ndigital.cpp, Line 722
Nth-pass file = C:\DOCUME~1\gen\LOCALS~1\Temp\\ice_age_2-tlrD_1080pa02716.bin

ERROR in file:E:\Sources\Current\Recode 2.2 Compare\Nero Recode II\burning\burn_ndigital.cpp, line:1884, error:Unspecified error (0x80004005)
12:09:24 File burn_ndigital.cpp, Line 200
NeroDigital: exit

[digigami]

And my best possible encoding at 3800 Kbps for libavcodec MPEG2 encoder
http://multimediacom.free.fr/Download/MPEG2-HQ5.m2v


Very nice Sagittaire. You have demonstrated, beyond a doubt, that MPEG-2 VBR has unlimited potential for HD-DVD and Blu-Ray.

X-mas today, I got 3 gifts for myself. 1) H.264 spec from ITU webstore. 2) NeroDigital (still waiting for license key) and 3) confirmation that VBR MPEG-2 is great for HD.

Thank you all!

PS.
svn co
./configure
make

x264 build succeeded.

[digigami]

Best possible encoding at 3800 Kbps for libavcodec MPEG2 encoder
http://multimediacom.free.fr/Download/MPEG2-HQ5.m2v


Now, I wait for someone to beat Sagittaire MPEG-2 VBR at 3800 with H.264 AVC - and no blocking artifacts.

Only small blocking in Sagittaire. Fade in to 20th Century Fox logo - difficult to see. His encode is almost perfect constant Q at 6, very high quality. Superbit DVD is 1-3 with very low Q-matrix. So at 3.8 mbits/s MPEG-2 VBR is very, very good for 1280x720x25p. At 10 mbits/s I think perfect quality is possible. Less than 1/3 of HD-DVD bitrate and 1/12 of Blu-Ray. MPEG-2 c'est bon, n'est pas?

Sagittaire - c'est champion. Avec custom matrix. C'est bon. J'attend pour H.264 at 3.8 mbits/s.
http://digigami.in-long-beach-ca.com/megapeg/Doom9/sagittaire-champion.png

[Sagittaire]

Now, I wait for someone to beat Sagittaire MPEG-2 VBR at 3800 with H.264 AVC - and no blocking artifacts.

But there are blocking artifacts on this MPEG2 encode (because average quant 6). You are really incredible ...

If I say H264 is better it's because I was already test the best MPEG2 encoder (IMO libavcodec) and the best H264 encoder (IMO Ateme HP).

The ratio quality is variable (source and level quality) but there are certainely something like 2.0 for ratio for this source at this quality level.

[IgorC]

Actually I agree with Sagitarrie. Libavcodec (enable in mecnoder) MPEG-2 provides best possbile for MPEG-2 quality. But it's still worse than H.264 High Profile.

[digigami]

I'm really happy now :)

Pour IgorC - la question d'accuracy de 'matices de quant'. Excuse-moi, mon francais est terrible.
-----
For IgorC - why I asked you questions about your system trying to determine the accuracy of the IDCT used by VLC.

This clip is from Sagittaire - Harry Potter trailer ripped from DVD, showing 4 different IDCT implementations (made png with Photoshop):
MPEG-2 decoder IDCT (http://local.digigami.com/mpressionist/screen.php?screen=idct-accuracy) accuracy comparison.
Notice AAN is very famous IDCT implementation, but compared to ChenSmith97, not so good for MPEG decoding. What IDCT does x264 use?

Windows Media Player is especially poor IDCT performance. VideoLAN is good. Hardware is?.... hardware dependent.

[digigami]

Actually I agree with Sagitarrie. Libavcodec (enable in mecnoder) MPEG-2 provides best possbile for MPEG-2 quality. But it's still worse than H.264 High Profile.

Yes, but a viable choice for HD-DVD and Blu-Ray which allow for 27 mbits/s MPEG-2 and 54 mbits/s MPEG-2 respectively.

J'attend pour H.264 AVC Ateme a 3.8 mbits/s.
----
I wait for a H.264 AVC Ateme at 3.8 mbits/s. Better yet, do one at 3.8 and one at 6 mbits/s.

[digigami]

Actually I agree with Sagitarrie. Libavcodec (enable in mecnoder) MPEG-2 provides best possbile for MPEG-2 quality. But it's still worse than H.264 High Profile.

Mais, MPEG-1 avez Q-2 stepsize seulement. Je pense que digigami avez une autre chance.

------
But, MPEG-1 has Q2 stepsize (in quantizer) only. I think that digigami has another chance. We have feature classification like libavcodec, but not I-frame injection (cut scene detection).

http://www.digigami.com/megapeg/blog/2005_03_16_archive.html