h.265 vs h.264

Random chit-chat and anything that doesn't belong elsewhere
Post Reply
rollin_eng
Veteran User
Posts: 4840
Joined: Wed May 04, 2011 11:06 pm

h.265 vs h.264

Post by rollin_eng »

tlindgren
Bright Spark User
Posts: 260
Joined: Sun May 03, 2009 2:14 pm

Re: h.265 vs h.264

Post by tlindgren »

No, it's complete nonsense! BBC R&D should be ashamed of posting that, but then the byline is by two of the report authors so there was no independent check.

They're testing using the h.264 and h.265 "reference coder" which are utter CRAP and not intended to be actually used for encoding, their mission is to generate samples of compliant bitstreams for testing. There's a number of other questionable things in it like using PSNR for the "objective evaluation", PSNR famously rewards blurring (!) which is why any serious test use something better.

It's well known that the h.264 reference code ("JM") actually manages to loose badly to XviD/DivX (old MPEG4 non-h.264 encoders!)... The h.265 reference code ("HM") isn't nowhere near as bad but it's not considered nor intended as an usable video encoder.

There are serious tests of video encoders, this isn't one of them despite (or perhaps because?) being published in an IEEE publication :D

The best video encoder test available right now is probably MSU which does yearly tests with actual usable encoders, invites all encoder authors to give feedback before, during and after and uses statistical methods that isn't long since discredited. The "full" version of the report costs money, but the free version is more than sufficient for anyone who isn't involved in actually writing an encoder.

Latest one was October 2015: http://compression.ru/video/codec_comparison/hevc_2015/
Overall Conclusions: Overall, the leaders in this comparison are x265, Intel MSS Hevc and x264!

Given that x264 has by far the best psycho-visual model of the three, I suspect a visual test would result in a win for x264 at higher bitrates, but x265/MSS now does win at really low bitrates. I know the x265 team is really busy with this so I expect them to finally beat x264 over the board in a year or two.

It's called a HEVC test, but includes four non-h.265 encoders including x264 (used as the "baseline") and VP9. VP9 didn't do too badly either, just not good enough to make it to the three-way tie at the top. And the psycho-visual side of VP9 is currently trailing x265 too.
nhyone
Bright Spark User
Posts: 252
Joined: Fri Jul 24, 2015 4:13 am

Re: h.265 vs h.264

Post by nhyone »

I find two main shortcomings with x265 today. There was a long standing blurring issue, which the developers deny exist. But in 1.8, they discovered the i64x64 caused it and hence it is disabled now. (I read this on Doom9.)

The second is that the presets are not fully tuned yet, unlike x264. Again, people have suggested custom settings that presumably work better than the presets in general. (The x265 psy is very conservative by default.)
Post Reply