Xiph.org has released Theora 1.1 alpha 1. This is the first release to include the rewritten and much improved "Thusnelda" encoder. See Xiph's press release for details; this mailing list post links to binaries.

i can confirm that windows binary outputs something (is theora VFR by default?).

edit: little test encode, you will need a nigtly firefox to see this;
http://somestuff.org/theora/ed.htm

Tusnelda might be much improved compared to the original (vp3-based?) encoder but the quality is still way below MPEG-4 (SP, ASP, AVC). Xiph really missed the boat on this one. In 2002 they got an agreement with On2 and it took Xiph foundation until the end of 2008 to get a 1.0 release. Six long years and no significant quality improvement. I have the greatest respect for the Xiph coders but in my opinion Theora is a huge failure and to be honest I don't even know why they bother with this severely outdated codec... I know I won't, maybe someone else will?

SP and ASP? Those are still a joke to me. It might be below AVC in terms of being able to scale to HD. Monty did a side by side comparison with H.264 and Thusnelda looked somewhat improved in terms of overall picture quality. I don't know that much about video so I really can't say I know for sure what was improved in the source code. I just know that macro block artifacts are nasty and not very pleasant on the eyes. The problem is only Monty is working on Theora with a small team of independent developers. I am not going to speculate on what Theora can fully achieve or what can be squeezed out the codec. It's not exactly a "huge failure". If it was Mozilla wouldn't be wasting $100,000 to fund development efforts for it.

http://web.mit.edu/xiphmont/Public/theora/demo.html

Monty mentions a "better" alternative similiar to what he had planned to do with Tarkin, but it would require breaking the specs.

As we all know, quality is subjective. I based my judgement on my own tests. Did you try out the new Tusnelda encoder? I did some testing and compared with both xvid and H.264. Needless to say, I got so depressed by the poor quality (even at really high bit-rates!) that I just stopped testing after 3-4 different video clips.

I've seen Monty's blog and progress reports and I was really looking forward to this release. I absolutely wish Theora would become an alternative to other codecs because I believe in open source, but from what I've seen it will probably never get there..

I know the small team hurts and not having any funding (besides some smaller donations here and there) is of course not good. I ask myself, where are all the Linux distributors? Where are Ubuntu, Novell & Red Hat? Where are all the companies who would benefit from a kick ass patent-free open source codec???

I downloaded it I am reluctant to encode with it though as I am upgrading to a machine that is faster for video transcoding. My machine has a very small cache and limited processing power. I am eager to test it it out though. Well you want to know what the ironic thing is? Monty actually works for Red Hat. The problem is the actual code is a bit outdated. It's impossible to implement any new features without breaking the spec!. They will either have to go that route to improve it or start working on a prototype of Tarkin from scratch with the funding that was given to them like they said they were going to! It's understandable why one would be a bit disappointed. A lot of people often forget, but there is also "Dirac" too which is a wavelet based codec designed with HD in mind!

it's not the goal to beat AVC, it just have to become good enough to be usefull... (i can't comment on the actual quality, since i'am obviously overlooking something with my cmd...)

I encoded this video to x264 and Xvid ( with extremely best/slowest settings) from lossless source http://media.xiph.org/ED/ to compare them to Theora.

x264 http://www.mediafire.com/?y3ozzntztjq
Xvid 1.3a (+VAQ patch) http://www.mediafire.com/?zbvekgdnyyn

You mention really high bit-rates but apparently Theora is not designed for really high bitrates. It seems it gives up visual 'transparency' in order to be an efficient low bitrate encoder.

I'm not an expert in either audio or video compression but I believe there may be an analogy to AAC+ which is designed for lower bitrates than standard AAC and the techniques it uses would be counterproductive at the high end.

You don't mention what sizes or uses you were encoding for, was it Blu-ray or more like Youtube web sized video?

"You're doing it wrong."

It sounds like you should be testing Dirac rather than Theora. Dirac is optimized for visually lossless operation at high bitrates.

At higher bitrates the 1.1 alpha has quality limits from using the old non-uniform quantizer matrices even at the higher bitrates and from the leaky forward DCT. Both of these should be resolved in the 1.1 release, but since they don't have much impact on 'webcasting bitrates' and they are tunings rather than infrastructure they've taken a lower priority. Even with these issues fixed, however, Theora has an upper limit which comes from the minimum quantizer step size in the format.

When speaking of quality for lossy video codecs you really do the world a disservice when you don't discuss bitrate in the same breath. Perhaps that should be a HA forum rule. If you don't care about bitrate, then use dirac in lossless mode and never worry about quality again.

The quality/bitrate curve isn't linear for any codec… some codecs work better at low rates, some at high rates, etc. A 40mbit/sec MPEG2 stream is going to look better than Theora (due to the above mentioned quantizer limitations and fDCT) ... but at 400kbit/sec Theora should be blowing away MPEG2. Theora is designed primarily for low bitrates, … the perfectly acceptable talking head/presentation at 150kbit/sec… general web video at a couple hundred kbit/sec. Decent, but not perfect quality at a megabit or two. If you're looking for something transparent, something archive grade... look to Dirac, as I mentioned. (And before you ask… Dirac does pretty poorly at low rates, clearly worse than Theora).

And also, lets be realistic: Theora is not going to beat MPEG-4 AVC, a more recent, more computationally expensive codec, which incorporates new features that cover the few things that Theora has over prior MPEG codecs (for example, in loop deblocking)… and AVC has non-trivial licensing costs. Perhaps the licensing doesn't matter to you, since you'll only be using video for private archival or such… but do recognize that if you discount the licensing you're discounting the principle advantage of Theora according to its maintainers, so you shouldn't be shocked if you find the format lacking.

Joe-random isn't as sensitive to quality as hydrogen audio users. There are list postings where people are declaring the new Theora to be a match for H.264. These people are nuts but they are also representative of the general public. Theora doesn't need to achieve the same bitrate/quality trade-off as H.264 for the licensing advantages to end up being more important… at least to people whom aren't audio/videophiles. It just needs to not completely suck… and thats a lot of what 1.1 is about.

If you think that it's regretful that people will tolerate less than perfect video quality… then please join in and aid the development of unencumbered audio/video codecs. As it stands today most of the public effort in codec development is going to projects like x264; propping up the quality of codecs owned by others.



Redhat employs Monty to work on this stuff. (Though RedHat has had him off on another project for the last few months, but he is working on Theora (and other Xiph projects) again now)

Now talking about Theora and other codecs. In this moment Theora has very unstable quality at high motion it makes this codec totally unusable. Yes, theora hasn't to be on par with H.264 but it's even worse than old ASP (Xvid). In two years when theora maybe will have acceptable quality for SD, but it will be already HD period and H.264_HQ_optimizied/H.265. Theora wil become useless again.

Years isn't fair if you're applying it to that particular complaint. In bitrate management mode the new encoder is pretty strict at keeping the instantaneous bitrate close to the target (this is important for streaming), and it only runs with one keyframe interval of buffer. As you and a few others have noted the current behaviour produces a periodic drop in quality near each keyframe. The new bitrate management mode is only a few weeks old now, clearly there are things to improve about it. (Though, compared to the old one its still worlds better, the old one would manage to totally miss the bitrate target in both directions *and* deliver terrible quality at the same time).

From a testing perspective if you are testing in bitrate managed mode the different codecs should be configured for the same buffer size in order to be fair… Small differences won't matter, but big differences can make ... a big difference. Unfortunately, everyone does management differently (and often the tools don't expose the right knobs), so getting actually comparable results can be hard. Streaming is an important use case… so people should be doing managed rate comparisons, … I don't really know what to suggest.

What was your vbv buffsize for the x264 encode?

why would that matter, if the target is download (that is some sort of 'generic' software player)?

edit: practical example, cheap AVC streaming via php
http://somestuff.org/flashAVC/flvplayer.ph...am-x640y352.flv
(i didn't bother with specifying the max VBV or something like that...)

edit2: about dirac, if you don't care about the size, then use huffyuv or lagarith or something (both lossless), these are the real FLACs of video world.

edit3: on the long run, the real winner here will be the one that will implement a simple (client side) system for p2p streaming of video (a good chance for Mozilla/Theora imho).

i don't think you need any 'best' settings (for x264) to beat theora at this point in time

2 years? HD over web is practially here, if i can do it;
http://somestuff.org/flashAVC/flvplayer.ph...d-x1280y720.mp4
anybody can....

If target is download, it wouldn't… except the posted Theora file was encoded with the equivalent of a buffsize of ≈1000. It's especially relevant because the quality problem IgorC spotted goes away completely when the encoder's instantaneous bitrate is not constrained.

Obviously Theora still needs to be improved … but it's a bit silly if we're holding Theora to a higher bitrate consistency standard then complaining about artifacts that stem directly from that constraint.



If you're *streaming* (and presumably thats what you're doing with the flv above) and you do not apply a virtual buffer constraint then one of two possible negative outcomes will happen: Either users which have perfectly adequate bandwidth will have their playback periodically halt and go into "buffering", or users will be forced to have an enormous pre-roll delay (say, tens of seconds for a several minute long clip) to be sure that playback probably doesn't halt. (The latter possibility becomes more and more inconvenient as the total clip length becomes longer)

Sure, it's possible that you're sending a 400kbit/sec video to a bunch of users who can all reliably stream 800kbit/sec, which diminishes (but does not remove) the need for buffer constraints… but if they all really have the capacity why not turn up your stream to 800 then apply a buffsize constraint to keep a 5 second buffer adequate for their needs?

As a web developer getting streaming buffering right is subtle, but it is an important factor in producing video that just works rather than something that works for the developer then falls on its face for actual users.

if it looks like streaming and behaves like streaming, what is it then? (actually it is so called 'fake' streaming in the example i posted... (or pseudo streaming))

ok, let's try and see if vbv actually makes any difference in this model,
what do you suggest are some good values for 800k stream?
will the 800k vbv limited stream look better or worse than 480k unlimited one?

I don't understand your LOL smiley. Is my analogy false or misleading? If so how?

NullC appears to be saying the same thing regarding Theora (but with actual knowledge of the technical details). Is it that my understanding that AAC plus is designed for lower bitrates is wrong?

The standards like MPEG ASP, H.263/+/++, H.264 are versatile. You can adjust settings efficiently for particular bitrate.
For example Xvid:
Low bitrates: H.263 quantization. High bitrates: MPEG or any other custom matrix. Some particular setting for min/max quantizers, etc.

H.264: in-loop deblocking filter. High/low values for low/high bitrates. Some psycovisual enhancements etc...

It's also applicable to use some external smart pre processing to increase compressibility. Some of them actually are very efficient.

Maybe on2 Vpx codecs can be considerated as AAC+ equivalent but it's rather to consider this situation as a lack of versatility as they're not better than MPEG versatile codecs at any bitrate. However Vpx codec also have adjustments for different bitrate but they're just more efficient for low bitrates than for high ones.

COOL.



Feel free to suggest some better codecs



Again

The problem is that Theora has just 1 or 2 developers at best ... and a very low priority for the FFMPEG/MPLAYER/MENCODER/VLC/other video developers, while not only decoding but also encoding of proprietary codecs has much more support, it took years until FFMPEG got Theora support included (see also: http://www.hydrogenaudio.org/forums/index....st&p=626231), and MENCODER still can't encode Theora at all.



Nice, but extremely slow, so not really a drop-in replacement for Theora. Also, the "upgrade" from a research codec to a useful codec is far away from done nowadays.

the only experience with dirac i have is with two guys showing me a bandwidth halver black-box (probably IBC in Amsterdam), but that was obviosly meant for in-house use and of course high data rates, anybody with some deeper experiences?

There is no difference between SD and HD. HD is just as 4 SD videos placed next to each other. Any codec good at SD will be good for HD and vice versa. JPEG has the same quality with 100x100px pictures as with 20 Mpx photographs.

1. not 4. Up to 6 (1920*1080)/(720*480).
2. Intra prediction of particular codec can be limited to SD resoltuion. So yes, there are some limited to SD videocodecs. Developers of x264 advise to test on HD sequences as many RD and PSY features are optimizied for high resolution and the way it's resized during degustation of encoded material.

3. Even if we discard points 1. and 2. there is still more important fact. Bandwidth (and it's main reason of my previous posts). Maybe Theora will be 'ok' for SD up to 1 Mbps. But then HD period is coming. Theora should have at least 3-4 Mbps for HD. While x264 is already good at (at least, even in future) 1.5x less bitrate.

4. I'm on side of open source and standards but considering slow development (Theora is already 7-8 years old ) and non optimistic views I don't see it's coming.

LWN has a good writeup on the subject: http://lwn.net/Articles/326697/ (many of the comments also contain details worth knowing). Monty's also given another status update, http://web.mit.edu/xiphmont/Public/theora/demo6.html. I don't understand the negativity on this thread- the new encoder's obviously much improved, will see considerable further improvement before it hits release (this is the first _alpha_ release, not a finished product), and AFAIK compares reasonably to other encoders at the kind of bitrates which one's likely to see on the web or on portable devices.

It's true that the returns in quality of increasing the bitrate well beyond that level diminish more quickly than for other codecs. HD vs SD is basically irrelevant- the question is bitrate not resolution, and if you move to HD resolutions and your bitrate doesn't increase by considerably more than the corresponding increase in resolution, then you're still looking at something which isn't at archival bitrates.

Theora needs Adaptive Quantization - ask Dark Shikari for it and Xvid rate control - it will be much better right away.

jensend: i would certainly prefer some (best of the best) theora example command lines, nothing to read there (just bunch of words).

p.s. and yes, we (as webmasters) got the point that we can use java applet until all browsers are <video> friendly...

There is another status update with recent improvements at Theora: Thusnelda project update 20090507, also comparing the PSNR with older Theora version and x264.

why would one plot psnr graphs and then negate them all the way (with words)?

I think the point was to advocate PSNR for comparing different version of the exact same codec on the same content, but not for comparing across different codecs or content (and even then be careful not to rely on it exclusively). The updated appendix discusses some additional context you would need to understand this, since it's rebutting a paper that wasn't otherwise mentioned in the original discussion.

Theora 1.1 alpha 2 was also released:
libtheora 1.1 alpha 2 announce
ffmpeg2theora Windows / Linux / OS X binaries compiled with libtheora 1.1 alpha 2

firefox 3.5 released.
http://firefox.com/

Theora 1.1.0 released
http://lists.xiph.org/pipermail/theora/200...ber/002807.html