Is the H.264 reference encoder the maximum quality limit?

[ChiDragon]

Just a theory question. Given infinite processor time, does the JM reference encoder produce the maximum quality possible for a H.264-compliant bitstream at a given profile/level/bitrate?

Put another way, is the goal of x264 and other implementations "only" to do things faster than the reference encoder software, or can they produce better quality as well?

[Dark Shikari]

JM is a very basic encoder implementation with typical algorithms and limited features. It's not "slow but good", it's slow and crap.

"Maximum quality" is impossible to achieve with any encoder -- even the theoretical approach of generating every single possible bitstream until the best result is found -- without a perfect measurement of what exactly quality is.

[ChiDragon]

Gotcha.

My question was driven by reading this study (http://bura.brunel.ac.uk/bitstream/2438/1181/1/Fullext.pdf) that concludes "It has been shown that the coding efficiency of 1080p/50 is very similar (simulations) or even better (subjective tests) than 1080i/25 despite the fact that twice the number of pixels have to be coded. This is due to the higher compression efficiency and better motion tracking of progressively scanned video signals compared to interlaced scanning."

They state "To avoid a bias due to encoder implementation preferences the H.264 software reference encoder JVT9.0 was used for the compression." I wondered if perhaps their results could still be biased towards progressive, for example if the reference encoder is just bad at encoding interlaced.

[Dark Shikari]

Probably about 99% of H.264 video compression papers use JM. Perhaps not coincidentally, 99% of H.264 video compression papers are a waste of trees.

[kypec]

Perhaps not coincidentally, 99% of H.264 video compression papers are a waste of trees.
What a relief that "papers" are mostly distributed in electronic form currently! :p

[kieranrk]

Having met a couple of the authors of that paper a couple of weeks ago, I would hope they use x264 for all their comparisons now.
To be fair, at the time in 2006, x264 wasn't well known and the only one they could really use was JM.

[oscord]

What is 1% ?
Thanks.