Confusion about Mpeg-4 and H264. Please help.

[acidsex]

Well, I tried searching the forum to understand better the differences between the two formats. I was fortunate enough to be a beta tester for Ateme H264 codec and honestly, I just dont see the benefit. Perhaps I am looking for the wrong things. Maybe someone can clarify this for me.

I use 3ivx for mpeg-4 compression and get decent quality for 700mb file sizes. However, when I use H264, for the same file size, I seem to get lesser quality. Why is that? I was under the impression that H264 was supposed to give me more quality and lesser file sizes. was I wrong in this line of thinking?

Add to it that the encoding speed seems to be unoptimized as of recent. I am excited to see what H264 is going to bring to our future but its just getting very hard to see the benefits of better quality and smaller file sizes.

Am I wrong on my assumptions?

[Valeron]

Because the current H.264 implement is "baby" only.
Any discussion about one codec with the early stage implement only is meaningless.

But IMO,even years later,I predict H.264 won't have too much advantage over the same period codec.
A great many codecs have been designed for far lower complexity than H.264(at least at decoding side,this conclusion is mathematic theory based),but very near or even higher quality(which lead to lower cost).And their's implement also will be improoved in the future.For example,On2 VP 6.2(I've no idea if VP7 and later version will be backward compatible with it),RV10's future,and AVS as well(a standard from China).Perhaps the only advantage of H.264 is an ISO/SMPTE/MPEGLA standard identification and the partical support from Hollywood and the broadcast industry.

Don't forget Microsoft is running their VC-1(aka WMV9)codec to a standard too.And it's also indicate to be far less complexity than H.264.

[thegeby]

Don't forget Microsoft is running their VC-1(aka WMV9)codec to a standard too.And it's also indicate to be far less complexity than H.264.

I wonder about this. I tested WMVHD clips against Ateme's H.264 at the same bitrate, with the Ateme transcoded from the WMV. In all cases the H.264 played with fewer hiatuses (hiati?)which would seem to indicate a lower playback load and less complexity.

Just a datapoint....

[Valeron]

Originally posted by thegeby
I wonder about this. I tested WMVHD clips against Ateme's H.264 at the same bitrate, with the Ateme transcoded from the WMV. In all cases the H.264 played with fewer hiatuses (hiati?)which would seem to indicate a lower playback load and less complexity.

Just a datapoint....

Different profile@level will lead to completely different result.
I wonder if Ateme's H.264 can decode so fast.Which profile@level do you apply when encoding?Is CABAC and B-frame enabled?
Finally,and the quality?
As far as I know,WMVHD is encode with WMV9 MP@HL(aka WMV9 Proffesional),which is the most complexity WMV9 profile.
And other words,this information is refer to a Microsoft's WMV9 introduce document,and conclusion based on both mathmatics discussion and a done DSP decoding speed research. The test compared WMV9 MP without B-frame with H.264 baseline on a DSP.

Bye.

[thegeby]

It was the Ateme H.264 with Cabac and B-frame, original 720p resolution at 6.5 Mb/s. In addition, H.264 permitted lowering the bitrate to 3-4 Mb/s with practically no deterioration (based on eyeballing)

[Valeron]

Originally posted by thegeby
It was the Ateme H.264 with Cabac and B-frame, original 720p resolution at 6.5 Mb/s. In addition, H.264 permitted lowering the bitrate to 3-4 Mb/s with practically no deterioration (based on eyeballing)

Great!That seems Ateme's H.264 comes with a very fast decoder.:)

BTW,have you test the clip at 3-4Mbps?
I'm not going to believe any codec can deliver acceptable quality at 720P with only 3-4Mbps for me(at least this moment).Especially for high motion clips.
(except Anime in RV10):D

[snacky]

I'm not going to believe any codec can deliver acceptable quality at 720P with only 3-4Mbps for me(at least this moment).

I'm afraid I don't recall where I nicked this graph from, but:

http://ikaruga.co.uk/~snacky/z/hi1.png

[thegeby]

Great!That seems Ateme's H.264 comes with a very fast decoder.

The decoder in both cases were bog standard Nero Showtime, running on a Celeron 2.5 Gig laptop

[bond]

till now i wasnt able to find any vc-1 specs on the net and i surely dont believe anything companies say about their codecs regarding quality, enc/decoding speed, complexity aso...

[Mr_Schizo]

Originally posted by bond
till now i wasnt able to find any vc-1 specs on the net and i surely dont believe anything companies say about their codecs regarding quality, enc/decoding speed, complexity aso...
The specs Draft is already available. Feel free to pm me or come back to #mpeg if you're interested to get a deeper look in the Draft.

[bobololo]

Originally posted by acidsex
Well, I tried searching the forum to understand better the differences between the two formats. I was fortunate enough to be a beta tester for Ateme H264 codec and honestly, I just dont see the benefit. Perhaps I am looking for the wrong things. Maybe someone can clarify this for me.

I use 3ivx for mpeg-4 compression and get decent quality for 700mb file sizes. However, when I use H264, for the same file size, I seem to get lesser quality. Why is that? I was under the impression that H264 was supposed to give me more quality and lesser file sizes. was I wrong in this line of thinking?

Add to it that the encoding speed seems to be unoptimized as of recent. I am excited to see what H264 is going to bring to our future but its just getting very hard to see the benefits of better quality and smaller file sizes.

Am I wrong on my assumptions?

Well H.264 is a new codec standard that benefits from enhanced coding tools compared to former standard. In H.264 you have tools like more efficient intra prediction, cabac, inloop deblocking filter, short transform, etc. All these new tools are supposed to provide a much better coding efficiency than mpeg-4 asp (3ivx).

Most of the testers from the beta reported that h.264 is actually better than older codecs in most cases. That's why your results are quite interesting and simply shows that we may still have issues to fix. Have you reported back your case in the quality feedback thread ? (providing your beta version, your settings, your source and maybe some sample where we can see the problem). Maybe you just spotted out a special case where our encoder performs badly ?

[bobololo]

Originally posted by Valeron
Because the current H.264 implement is "baby" only.
Any discussion about one codec with the early stage implement only is meaningless.
You should have a look at this thread (http://forum.doom9.org/showthread.php?s=&threadid=82036). It may change you mind ;)

[helix]

Just be glad you acually got to beta test. I want to :(

[acidsex]

I checked out some of the other clips on the other thread posted above and I must say, it is amazing. personally, I think having good audio makes the viewing ten times better.

Thats the main snag I have hit currently. How do I get my HE-AAC audio muxed with my H264 output into an .mp4? Is there a specific tool I should be using?

[CruNcher]

H.264 permitted lowering the bitrate to 3-4 Mb/s with practically no deterioration (based on eyeballing)


as you can see from my samples i posted in the quality and feedback thread there is a difference in the preservation of high frequency film grain, it gets almost completly decimated (washed into nearby pixels) couseing blinking blocks in H264 at the moment, but thats still a thing thats worked on see the posts by bobololo about that subject :)

[Valeron]

Originally posted by bobololo
You should have a look at this thread (http://forum.doom9.org/showthread.php?s=&threadid=82036). It may change you mind ;)

I've paid attention to Ateme's H.264 for several days,according to yours disscusion,that seems a great H.264 implementation.
But where can I get the beta??

I can't found it on the site of Ateme.

Thanks.

[bobololo]

Originally posted by acidsex
I checked out some of the other clips on the other thread posted above and I must say, it is amazing. personally, I think having good audio makes the viewing ten times better.

Thats the main snag I have hit currently. How do I get my HE-AAC audio muxed with my H264 output into an .mp4? Is there a specific tool I should be using?

In the beta packages, you have mp4muxer.exe for this purpose.

[bobololo]

@helix & Valeron: The beta test is over. You have to wait for the final release now.

[helix]

Oh... Well, as long as one's comming, i'm happy. :D

[saratoga]

Originally posted by bobololo
Well H.264 is a new codec standard that benefits from enhanced coding tools compared to former standard. In H.264 you have tools like more efficient intra prediction, cabac, inloop deblocking filter, short transform, etc. All these new tools are supposed to provide a much better coding efficiency than mpeg-4 asp (3ivx).



3ivx is actually MPEG4-SP at the moment. I think theres supposed to be an ASP version in the works, but the stuff on their website doesn't do ASP yet AFAIK.

[stephanV]

3ivx has ASP features like the MPEG Q-matrix...

[saratoga]

I didn't know that was an ASP feature. SP is limited to H.263 then?

[stephanV]

read here (http://forum.doom9.org/showthread.php?s=&threadid=73022)

:)

[hworldjj]

If you want to verify the quality of H.264, pls use JM from http://bs.hhi.de/~suehring/tml/ instead of other commercial codecs such as nero, vss, etc, which are just partial implementation.

[skal]

Originally posted by snacky
I'm afraid I don't recall where I nicked this graph from, but:

http://ikaruga.co.uk/~snacky/z/hi1.png


Snaky,

this graph is from this (rather good) article:

http://fastvdo.com/spie04/spie04-h264OverviewPaper.pdf

... which brings us to the original topic ;)

Skal

[bobololo]

Originally posted by hworldjj
If you want to verify the quality of H.264, pls use JM from http://bs.hhi.de/~suehring/tml/ instead of other commercial codecs such as nero, vss, etc, which are just partial implementation.

Well even with JM it's hard to evaluate since it still lacks a good rate control which can bring significant improvement to the quality.

[hworldjj]

Yes, rate control is important for any video codec, but we still can compare the quality of them. A fixed quant can avoid the influence of it. Even though the quant range is different for mpeg-4 and H.264, we can chose two reasonable quants to evaluate them.

Or another way, just compare xvid and jm8.6 with their individual rate control, you can see the difference if only you set the right parameters for them.