I'm not worried about there being a ton of later comparisons as Theora continues to improve :-)

In general, we recommend people not use non-default settings with our codecs. Unlike many other projects, we put effort into making sure the defaults are correct.

That said, this is one case where the Theora default was suboptimal because it was the wrong use case. We need to reword the way these options are specified. For bitrate manged modes, there's two common ways to use them: Very tightly constrained rate for a fixed-rate channel (like ISDN or low-rate DSL) and a more relaxed management that is simply trying to hit some size over the space of the whole movie (usually a two-pass mode, fit-a-DVD-on-a-CD type use). Right now, the Theora tools all do the first and that's how your test was performed. The h264 was encoded with the second. So that's a penalty to the constrained encoder.

VBR vs VBR is a more accurate direct test of encoder capability. Don't get me wrong, x264 is going to win that test too, just not by as much.

Only one point I wanted to mention (since the article and comments have all been--- oddly balanced for Slashdot)

The article points out that current Thusnelda is not as high quality as the best available h264 encoder at high bitrate video and unlimited encoding time. No argument there, it's true. Thusnelda still has a ways to go, despite the distance it's come; the current alpha still has no Adaptive Quant whatsoever, which will go in before final release.

However, the vast majority of users are not using x264. If you look at the h264 YouTube encoder, which has been designed for speed rather than 'work as long as you like to optimize the output', suddenly Theora is exactly on-par. In short--- Theora is every bit as good as the way that the real world is going to end up using h264 for the forseeable future. And the users of that 'inferior' h264 encoder seem pretty happy with it.

Anyway, this isn't disagreeing with anything you've said, it's simply a practical way to look at the difference.

Monty

Uh, that version is also six months old. We're talking about *new*, remember? Not last thanksgiving before any version of Thusnelda was even released.

"Ogg" is actually a term from an early internet game.

http://en.wikipedia.org/wiki/Ogging

Theora is named after Theora Jones, a secondary heroine character from the movie 'Max Headroom: 20 Minutes into the Future' about a dystopian future where video media is overwhelming, centralized, oppressive, dangerous, and an off switch on a television is illegal:

http://en.wikipedia.org/wiki/Theora_Jones#Theora_Jones

"Xiph" is actually from the Greek ξÎÏÎÏ (sword) by way of 'Xiphophorus' (sword-bearing, pseudolatin?) from the genus name of a fish (Xiphophorus helleri). Which is where I picked it up in middle school. I'd been using it for my software projects since I was 14 or so and by the time Xiph.Org was a real thing [many many years later] I wanted to change the name to something less silly and my co-founders voted me down. They liked Xiph. It became the precedent-setting silly name.

Vorbis is from Terry Pratchett's _Small Gods_ and I dearly hope Mr. Pratchett considers it a compliment. It was meant as tribute to my favorite fictional villain, Archdeacon Vorbis. "A mind like a steel marble"

Well, one tester (and Greg's graph was generated to rebut his graph) was converting each output frame to PNG and then feeding them into one of a number of PSNR tools one by one to get a PSNR result. The conversion from YCb'Cr' to RGB is lossy, but apparently this particular author didn't know that.

He was also using multiple PSNR tools because some were mysteriously not working with some video inputs. Given that there's no one standardized way to calculate PSNR, that led to additional data lulz.

And for x264, he apparently didn't generate his own numbers, he just used someone else's published numbers.

Anyhow, He reported that x264 was 30-ishdB (!) better than Theora. Wha? If every Theora frame was black, that still wouldn't account for that much difference. YUV12 is only 40-45dB deep!

In other words, the whole point of the graph was originally to illustrate and rebut these errors, and it turned out to be a nice regression test too.

Also, for the record, the x264 curve is not perfectly smooth, but that it's as smooth as it is attests to the fact that it's a nicely tuned codec. That Theora is lumpier is one indication it still has more tuning to be done.

Also, Greg's response below is way more levelheaded than mine. He actually collected the data himself (so has more detailed, accurate and first-hand knowledge) and he also probably hasn't been drinking whiskey all night.

Uh, this article here on Slashdot is about *brand new* improvements to Theora, how it is vastly better from original Theora of even one year ago, and also about how a really old broken version of ffmpeg was also causing really terrible quality problems....

So you post several year old screenshots made with an old, unknown [but definitely broken due to age] version of ffmpeg.

BRILLIANT. Seriously, dude, fewer bonghits.

You have to measure the PSNR of each codec with the same tool, silly (and avoid doing colorspace conversions which are lossy in the interchange. Keep the output in YCb'Cr' format). If you're using the x264 encoder's reported PSNR *cough*ahem* it's known to be wrong. It always reports way higher than other tools, like it's forgetting chroma is subsampled or its log-space algebra is just wrong or something.

Let me check myself with the clip linked in the article....mmmm lessee.... yep! that's what you're doing. So, BZZZT, no gold star, try again.

To be fair, the whole thing is part of 'the Ogg Project'. Saying 'Theora is Ogg' is not actually incorrect, and it might get you laid at parties.

No need to be such a stickler, here have a beer.

That practice is called 'dumping' and is illegal for most goods and services, at least in the United States.

Why would either Slashdot or Streisand apply here? Just wondering. Slashdot traffic isn't exactly likely to bother MITnet much, and I'm hardly likely to censor my own post or have MIT remove it.

I *had* been thinking of adding some gradient improvement screencap examples though.

And following Thusnelda will be Snuppflog. They're just internal project names.

Intel chooses boring internal codenames like towns, we choose silly things that our incredulous detractors dare us to use. But only if we like them.

Actually, I oversimplified to the point of being partly misleading--- the few famous lab studies conducted on DDT that implicated it in egg shell thinning and carcinogenesis turned out to have been seriously flawed. When verifying the study results, others later found that in some studies the DDT had been contaminated with PCBs (powerful toxins/mutagens), in another rat study found the feed supply to have been heavily contaminated with aflotoxin (also a carcinogen) due to moldy conditions, etc. The studies that had originally implicated DDT were not in the majority, and so science was skeptical even at the time that DDT was the culprit (not that the press cared). When the flaws in the experiments were found and the studies repeated, they did not find DDT to have any significant affect on animal health or the reproductive success of bids (ie, no egg shell thinning). It was thus posited (but not proved) that if DDT was having an effect in wild populations, it was due to industrial contamination of the DDT, not the DDT itself.

Science corrected its mistake, but the media thought the original story was way more interesting. Truthiness isn't a new concept.

It was found out some time ago that it wasn't DDT itself thinning the shells of birds' eggs such that the eggs were too fragile to be viable, it was contaminants and impurities in the DDT due to the way it was being manufactured. DDT itself proved relatively innocuous, but this was established long after it had already been banned.

If it can be manufactured without the toxic impurities, it is indeed a miracle cure for many of the developing world's ills.

The most important question unasked

Cheers,
Monty
Xiph.Org