Follow-up on Microsoft's H.264 choice for IE9 | Hacker News Reader

Follow-up on Microsoft's H.264 choice for IE9

blogs.msdn.com

21 pointshackermom34 comments

Here's something the average developer may not realize: Theora just looks starkly bad compared to H.264 at web bitrates. I didn't use to think this, when all I had seen was some comparison articles with objective measures and still images but then I actually tried encoding them myself and playing them simultaneously.

For those genuinely curious for discovery (rather than making rhetoric) try this html in Chrome or a browser that supports both H.264 and Theora. It's an example where someone has encoded a video in both formats at the same bitrate. (I'm not going to link or host it because I don't want to kill the site the video is hosted on.)

    <html><head></head><body>
        <video controls="" autoplay="" type="video/ogg" src="http://www.doceo.com/SD1_1.ogv"></video>
        <video controls="" autoplay="" type="video/mp4" src="http://www.doceo.com/SD.mp4"></video>
    </body>

So try that, and then after watching that try to tell me that the difference in visual quality is slight. After doing this myself (those videos are not mine but my results were similar) I can only conclude that supporters of Theora as a pragmatic choice (as opposed to an idealistic or emotional one) for delivering professional video content have never done such a test.

What encoding parameters did you use?

I tried very hard to use ogg for encoding a video about 1.5 years ago as well (it was a requirement) and it was so painful. The tools are awful and it took me 5 tries just to get something that wouldn't crash most players if you tried seeking.

After I finally got something valid, I ended up settling with a file 1.6x the size of the h264 file I would've created that looked only a little worse.

For amateurs like me ogg is hard. h264 is easy.

ffmpeg2theora is easy. http://v2v.cc/~j/ffmpeg2theora/

Ok, command line interfaces are certainly no problem for anyone reading HN but calling them easy seems slightly, uhm, naive.

Also, manpages are awful if all you want to do is encode easily (they also aren’t such a nice read) and their only other documentation – the examples pages – is chuck full with broken links.

The post I replied to sounded like a programmer's task, so cli is no problem there.

There are not many important things on the manpage. Usually you would set video quality (-v) and maybe audio (-a) and that's it. Theora does not offer a gazillion switches, it is an easy codec.

I am indeed a programmer and command line doesn't faze me. I use both ffmpeg and mencoder pretty often and I think ffmpeg2theora might've been the one that actually produced something that wasn't crashing players like mad. But even for me it took many tries to get something usable. My younger brother who's web savvy but not a programmer would not have been able to do it without coaching. My mom would've been a hopeless case.

Saying that ffmpeg2theora is easy ignores the huge portion of the population who create their movie in Adobe Premiere or iMovie or whatever and click export and choose "TV" or "Youtube" or whatever format with automatic defaults and it looks good and it works. Until you get a GUI on Windows that's even better than Super (which I really like but is really scary to people who don't do much video encoding) it won't get anywhere. Handbrake is the only program I recommend to non-coders. (And my brother uses it fine.)

Handbrake has a horrendous UI. It just about copes if you need to rip a DVD to the best possible quality, but if you stray from that happy path it's murder.

Have you tried the Miro Video Converter? It's supposed to be easy and does H.264 for iPod, Nexus 1 etc. as well as Theora at web rates:

http://www.mirovideoconverter.com/

No, I'd never seen that. It looks like exactly what I've wanted to recommend to people.

I didn't say I was happy with recommending handbrake, it was simply the least complicated option I had seen and my brother seemed to be able to muddle through so it was just good enough.

Miro looks a lot better though, thanks!

Just to correct myself slightly it appear to be (currently) using MPEG4 part 2 (Xvid) for most of its presets, though they tested H.264 encodes for certain phones.

https://develop.participatoryculture.org/trac/democracy/wiki...

I'd be interested in their reasoning for this, though I remember for a while the best quality possible on iPods regardless of time to encode was high bitrate MPEG4. I wonder how big a difference there is at these screen sizes?

What version of Theora did you use? What encoder? What did you use for the H264 one. Can you please provide the source video file?

There are http://people.xiph.org/~maikmerten/youtube/ and http://people.xiph.org/~greg/video/ytcompare/comparison.html , did you see them?

edit: Above is evil. Very evil. Or stupid. I encoded the audio from the MP4 to Vorbis to a size of ~800 Kilobytes and it sounds good. Also the Theora video is so bad and damaged, it does not remotely look like a normal encode. I would like to know what went wrong there. Anyways, disgard parent post and the links inside. It's nothing but FUD in my eyes (and ears).

edit2: I also get far better visuals than in the linked Theora by even transcoding linked H264 with "-K 170 -V 465" in ffmpeg2theora...

I also got better visuals by transcoding from the MP4 again, so something is very messed up with Doceo's Theora encode. One-pass with 450 kbps video, for about the same size file. (I didn't test the audio).

The tests performed on Xiph.org (and all other locations bringing up the same topic) are, in plain words, downright shitty. They are not using any advanced or even intermediate features for the H.264 encoder in question, and I believe they are not even using x264 (which happens to be the best H.264 encoder available), but instead one of the other "underperformers" available, and indirectly (or perhaps even deliberately) spreading the misnomer that video always looks the same at N kbps bitrate, while the truth is that the result can be higly, highly different depending on the feature set used in the video stream, as well as the feature set used to analyze frame by frame.

In a test like this, it's anything but fair to use the first, simplest h.264 encoder at hand, with the weakest settings possible, and then claim that "this is what h.264 looks like", when the most substantial component in video quality is not the data format itself, but the frame analyzer software and the stream programmer (read: the encoder itself). I've dealt in video ripping and encoding since the late 90s, having experience in pretty much every mpeg-1, mpeg-2, mpeg-4 asp and mpeg-4/avc encoder available, and Theora JUST DOES NOT COME CLOSE to what x264 can create for the h.264 format. In fact, it barely beats what Xvid can produce for the older mpeg-4 asp format.

So far I have not come across a single, proper, honorable comparison of h.264's capabilites compared to Theora. One of these damned days I hope I myself will be bothered enough to get my thumbs out of my ass and put the definite facts out there for people to see.

add.: if anyone wonders, i have ripped and encoded video (and audio) for illegal distribution in the so-called "audio/video scene" for almost 13 years now

The Xiph comparisons were made with x264 when possible. It would have been silly to use something different than Baseline profile as Theora cannot compete with H264 on an advanced "HQ" level and no-one ever seriously said so. We are talking about web videos, something where Baseline is the main target.

Just for the reference: you actually don't need to use any "high profile" features of h.264 to have x264 outperform Theora. In fact, you don't even have to max out the baseline profile features.

I did not say that Theora looks better or exactly as good as H264. I know how good H264 is. But for many uses the differences are marginally and do not justify using a non-free codec in my opinion.

If you would make your own fair comparison (article/site), it would be much appreciated.

Just for further reference: you don't need to use any "high profile" features of H.264 for x264 to outperform most other H.264 implementations (Adobe's, Apple's etc.) when they are using all the high profile features available to them.

I find it weird to use a GPL encoder that's had a lot of optimisation work done on it, to undermine an open format where the encoder hasn't had a lot of work done on it. If the format is the key rather than the implementation, then why are so many of them mediocre?

The Xiph.org test linked to isn't Theora vs H.264, it's Theora vs. Youtube. It's made abundantly clear in the text that this is the case, that they were responding to a claim made by a Google employee and they successfully proved that claim to silly.

People seem to get really annoyed that Xiph aren't making whatever argument they wanted made, c'est la vie.

Here's a comment about that comparison from an x264 dev:

"We talked about that on IRC, both with x264 and Theora people: all considered it one of the worst articles they had ever seen.

They screwed up aspect ratio, luma levels, BT.601 vs BT.709, and so many other things in their comparison that you might as well throw the entire thing out. Plus they used an incredibly bad H.264 encoder. It’s practically a checklist of how to be completely incompetent idiots."

http://x264dev.multimedia.cx/?p=292#comment-2768

You couldn't have picked a worse test, out of the many that have been done over the last year or so.

The fundamental mistake he's made (though there's a whole catalog of them) is that the bitrate for Theora is constant, while the H.264 is variable so H.264 saves up bits from the easy parts for the hard parts while Theora is set up as if it was live streaming and so it just uses the same bandwidth for both the easy and hard bits.

The original article that the two clips are from is here, you can see responses from Xiph and others pointing out what he's doing wrong, though it reads strangely as he had to rewrite it twice due to feedback before he gave up:

http://www.streaminglearningcenter.com/articles/ogg-vs-h264-...

"We are aware that this commitment is set to expire in 2016, but fully expect to commit to supporting the extension of this license and associated terms beyond that date."

They're going to ask a profit-maximizing corporation to not raise prices after they've got the entire globe locked-in? Sounds like another great plan from the boys in redmond.

Why not ask them to commit to no fees for 10 or 15 years today? What extra leverage do they think they'll have after they commit to H.264 alone for 5 years?

Every new codec has had lower fees than the previous generation; by 2015 we should have H.265 which will probably have free streaming as a teaser. Maybe H.264 will be forced to keep its fees lower than the switching cost to H.265.

Developers who want to use the same markup today across different browsers rely on plug-ins.

I'm honestly not sure what they mean by this, either the writer has not expressed his thoughts clearly or there is a severe issue here.

Developers who want to use the same markup across different browsers rely on standards. Each browser supports a sub-set of the complete HTML, JS & CSS standards and the technologies that developers use is, for the most part, the lowest common denominator of that subset for the browsers they are targeting.

Plugins come into their own when you are trying to do something that falls outside the capabilities of the lowest common denominator (Silverlight & Flash being good examples), but they should not be relied upon for the 'bread & butter' of the web, HTML & CSS.

Why not just make an API and let the community add the codec?

//Anders

Because then you get sites that say "to view this free porn, just install our custom codec", and of course the codec is malware.

Silverlight 3 does that: http://www.silverlight.net/getstarted/silverlight3/ "With the new Raw AV pipeline, Silverlight can easily support a wide variety of third-party codecs. Audio and video can be decoded outside the runtime and rendered in Silverlight, extending format support beyond the native codecs. "

> "Third-party applications that simply make calls to the H.264 code in Windows (and which do not incorporate any H.264 code directly) are covered by Microsoft’s license of H.264."

Seems like to me one could at least write a Windows only extension for Firefox that would allow Firefox to easily view H.264 without having to worry about royalties if they use of the code in Windows.

Yes, but then Firefox users on Mac OS X or Linux would be out of luck.

I believe Apple has the same deal.

How much would one have to charge each user to release a library for Linux that covers the H.264 fee?

http://www.mpegla.com/main/programs/avc/Documents/AVC_TermsS...

20 cents per user at most.

IIRC H.264 decode + AAC decode is less than $2. Linux users still wouldn't pay it, though. Fluendo charges significantly more, but they also have pretty low volume.

Also, every ATI or NVIDIA graphics card should already have a licensed decoder in it.

Flash already covers this, doesn't it?

After watching the BS IE team spread about Chrome, they have lost whatever respect I had for them for trying to make IE a better browser.

http://geektechnica.com/2010/04/microsoft-continues-its-trad...

Its just a matter of time before their market share drops down to tens. Unfortunately they will never completely go away.

The article you linked is weak. The first retort isn't much of a retort really, over 90% of chrome users don't change their search engine and Google does see everything you type into what normal people used to consider to be address bar. You can call it all kinds of fancy names, but midnight visits to goatfetish.com that used to be private are now google's property.

My midnight visits to goatfetish.com aren't private?!

Reply on news.ycombinator.com