Theora: Thusnelda project update 20090507

[IgorC]

I think this PSNR graph isn't representative.
First of all resolution was QCIF 176 × 144
Second Akiyo sample is very low motion video and consequently very easy to compress.
From my experience Theora has big problem with high motion. Quality is droped too much.

Today Theora isn't on par even with Xvid. Let alone x264.

http://www.hydrogenaudio.org/forums/...dpost&p=625394

[benwaggoner]

Originally Posted by Selur

Not if you look at it from a video encoding forum perspective, people here like codecs that can give a good quality/size ratio.

Also, while I've heard plenty of assertions that Theora is patent free, I've not actually heard anyone who's an expert on video patent IP validate those assertions.

That aside, there's a real cost advantage from improved compression. Any publisher picking a codec would compare the H.264 license fee for them versus the incrased bandwidth costs and reduced audience capable of hitting a minimum quality level. Plus compare the installed base of platforms/devices with the two decoders.

I don't have any stake in this came myself; Silverlight's my baby, and with Raw AV we'd be able to host Ogg + Theora + Vorbis as well as anything else..

But if patent status was that compelling, why not MPEG-1 or H.263? It seems like there are other codecs that are more clearly free-to-implement and have much more refined implementations than Theora or Dirac. Or at least using H.263 as a base and innovating from there.

And I haven't heard much to suggest that Theora could substantially outperform a well-tuned H.263 (MPEG-4 pt Short Header after all...), and we actually have well-tuned H.263 encoders. How's xvid in Short Header mode?

[Dark Shikari]

Originally Posted by benwaggoner

Also, while I've heard plenty of assertions that Theora is patent free, I've not actually heard anyone who's an expert on video patent IP validate those assertions.

I'd personally say Theora is worse-off patent-wise than H.264.

H.264 is the the devil we know. We know exactly what patents cover it because all the companies involved were required to submit their IP as part of the process. There aren't any surprises.

Theora is the devil we don't know. Nobody has any idea what patents it covers. Nobody has done a real analysis. All we have is Xiph's word that it doesn't violate anything, which is pretty much meaningless.

[dragsidious]

But if patent status was that compelling, why not MPEG-1 or H.263? It seems like there are other codecs that are more clearly free-to-implement and have much more refined implementations than Theora or Dirac. Or at least using H.263 as a base and innovating from there.

I am not that familar with the patents around codecs, but I think that H.263 is still patent encumbered. MPEG-1 layer 3 audio is certainly still patent protected and while _legal_ open source implimentations exist (the codecs from Fluendo) they are still subject to redistribution restrictions because of the patent licensing. Although Mpeg-1 is unemcumbered and it's patents are expired.

As for Theora vs Mpeg-1 for streaming internet video.. I don't know which one is better, but I suspect Theora is.

I'd personally say Theora is worse-off patent-wise than H.264.

H.264 is the the devil we know. We know exactly what patents cover it because all the companies involved were required to submit their IP as part of the process. There aren't any surprises.

That is a useful viewpoint for a lot of situations, but it's not really valid in this one.

Remember the goal here is to create something that people can redistribute and modify without restrictions. That is hack it, patch it, sell it, bundle it with consumer goods and services, use it in a professional capacity etc etc. If you try to do that with a H.264 codec then you will have a 100% chance of violating patent law. With Theora you can still use it, right now, in that manner legally.

Nobody has stepped forward yet with claims.

The problems with doing patent searches is that you can never be 100% sure what the patents actually mean. Even if you hire a lawyer who knew what they are doing you can't be sure. There are literally hundreds of new software patents each month and many are not limited in scope to specific programs... that is it's quite possible that a method of, say, compressing text in a database for heuristic searches could apply to a video codec. With literially tens of thousands of new software patents every year the chances of you violating somebody elses patents somewhere, by complete accident, is about 100% for any non-trivial program.

Also the fact that people are pushing patents licenses on H.264 does not isolate you in any way from patent lawsuites coming from other parties. Microsoft has learned that a couple times with regards to MP3...

And there is a trick here:

If you are show in court to violate a patent on purpose your pentalties can be tripled. So if you do end up doing patent audits and later on a court case shows that you do, indeed, violate a patent then that patent search/audit can come back to burn you. It will be easy for the other side to show you did compare the patent against your code and you still violated the patent... so it wouldn't be hard to convince a judge that you did it on purpose and was simply hoping not to get caught. So the general approach that has been recommended to the open source crowd by lawyers is to don't violate patents that people are publicly enforcing, but don't go fishing for patents to avoid either... patent audits and searches will have nearly zero benefit.

So in this case, were you want to have something people can modify, use, and redistribute freely, then yes not knowing you violate licensing is much much much better then knowingly violating patents, and that is the difference between something like Theora vs H.264


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

Keep in mind that in my personal life, the things I use and create... pfff I've never used Theora for anything. Divx-style or H.264 all-the-way. I've ripped movies for copying them to my cell phone or using on my netbook and I am always trying to find a nice balance between speed, efficiency, compatibility, and file sizes. The avialable tools for Theora simply don't match what I can get withj other open source stuff.

But if I was to recommend doing something like realtime internet streaming to my place at work, for some personal commercial gain, or some non-profit (etc) and we couldn't afford the licensing fees (or in the case for H.264, future licensing fees) then Theora would be the best choice.

Of course that is in the USA. In other counties they are not quite so patent insane so it varies.


I am not a lawyer, obviously. (thank god)

[Dark Shikari]

Originally Posted by dragsidious

Remember the goal here is to create something that people can redistribute and modify without restrictions. That is hack it, patch it, sell it, bundle it with consumer goods and services, use it in a professional capacity etc etc. If you try to do that with a H.264 codec then you will have a 100% chance of violating patent law. With Theora you can still use it, right now, in that manner legally.

Nobody has stepped forward yet with claims.

But while Theora is the devil we don't know, and H.264 is the devil we know, MPEG-1 is out of patent period and is guaranteed to be safe.

[benwaggoner]

Originally Posted by Dark Shikari

But while Theora is the devil we don't know, and H.264 is the devil we know, MPEG-1 is out of patent period and is guaranteed to be safe.

Anyone done a RD curve against modern MPEG-1 implementations for webish bitrates? I wonder how it would do coupled with a decent postprocessing filter.

Back when I did the Grolier's CD-ROM encyclopedia encodes in MPEG-1 (96-98?), I was able to do pretty good looking 320x240p24 down around 800 Kbps VBR with Heuris MPEG Power Professional. And that was not a good encoder compared to what came in later years once the DVD boom kicked in.

I wonder how it would compare to current Theora builds. Theora certainly has a lot more low-hanging fruit, but if the goal is something unambiguously free to implement...

[benwaggoner]

Originally Posted by benwaggoner

Anyone done a RD curve against modern MPEG-1 implementations for webish bitrates? I wonder how it would do coupled with a decent postprocessing filter...

Oh, how soon we forget. I just did a test encode, and MPEG-1 and MP2 are still crap at any reasonable web bitrate, even with 2-pass VBR etcetera.

[Shapierian]

Originally Posted by Dark Shikari

But while Theora is the devil we don't know, and H.264 is the devil we know, MPEG-1 is out of patent period and is guaranteed to be safe.

Even MPEG-1 is still susceptible to yet unseen submarine patents filed under the old pre-1995 submarine friendly rules. If course such a submarine would most likely also torpedo any other codec.