Slashdot Mirror
BBC Confirms 50% Bitrate Savings For H.265/HEVC Vs H.264/AVC (bbc.co.uk)
An anonymous reader writes: A research team from the BBC has done a series of tests to confirm earlier computations showing a ~50% savings in bit rate for H.265/HEVC compared to video using H.264/AVC at comparable quality. "The subjective tests used a carefully selected set of coded video sequences at four different picture sizes: UHD (3840x2160 and 4096x2048), 1080p (1920x1080), 720p (1280x720) and 480p (832x480), at frame rates of 30Hz, 50Hz, or 60Hz. The video content was chosen to represent diverse spatial and temporal characteristics, and then coded using HEVC and AVC standards at a wide span of bit rates producing a variety of quality levels." Here is the full published analysis. "The tests confirmed the significant compression efficiency improvements achieved in HEVC, verifying the results previously reported using objective quality metrics (PSNR based methods)." The team did not test against VP9, which is shaping up to be an impressive standard as well.
I've played in this space in a former position. Interesting lessons learned:
- PSNR is nearly worthless: An image with almost the same score can look terrible. Not all the time, but enough of the time.
- The only quantitative test I found that worked reliably was an old analog Tektronix PQA500 (lots of work to use for digital CODEC.)
- Management didn't like the PQA data (it said our product was terrible), decided to use PSNR data (product is great!)
- Customers fixed this discrepancy and product line failed spectacularly (due to video quality, surprise!)
- I never could find any published information sufficient to recreate the Tek PQA algorithm.
Name another media company that went out of their way to develop a patent-free media codec that was independent and competitive with other codecs of the time? (Google Dirac)
The BBC are publicly-funded, and under immense pressure to justify their funding at the moment - there's talk of scrapping the TV licence, and with it the BBC. They receive no advertising revenue in the UK at all. They only get some foreign revenue from sale of media (not even their own codecs or patents), and that goes to their commercial arm which isn't funding stuff like this.
There's no profit in them evaluating codecs, only if they then go out and build their own hardware that uses it. They didn't manage to do that with Dirac either, so why they would with this I have no idea.
All they want to know is what's best to push through iPlayer and store in their archive.
I have done my own comparisons of AVC (using x264, single-thread, veryslow preset) and HEVC (using x265, disabling wavefront processing because it slightly reduces quality, veryslow preset). All 1080p video, significant because HEVC is supposed to scale to 4K better than AVC.
My conclusions:
1) x265 takes FAR longer to encode, but we knew that. Understandable.
2) When "low in bits", x265 blurs images rather than making them look blocky. This sometimes looks better but to me often looks worse.
3) x265 seems to force a denoise filter. Video is far easier to encode efficiently when denoised, so I figure this is part of the data savings. It's a bit of a cheat, however, because I can get far smaller file sizes by running a denoise filter myself for x264-encoded video.
I looked closely, for example, at Captain America the Blu-ray. Much of the detail of, e.g. car leather and grass and tree leaves is lost in an x265 encode, even at about the same overall data rate as x265/
x265 supports "--tune grain", roughly analogous to "--tune film" for x264, but it makes the video vastly larger -- often larger than x264's version, and it often looks worse. It does a better job of keeping grain, however.
My experience is very similar to many others' in forums. I had committed to switching my encoding to HEVC, but the results of my tests showed it is not ready for prime time. Some may not mind blurry ("soft" is probably a better word) video, or video that looks like it has been through a denoise filter, but I do.
This is not to say that x265 is junk. I am sure it will mature over time just like x264 had to over time. x264 started out as being not all that much better than divx, the previous generation.
Computer Science is no more about computers than astronomy is about telescopes. --E. W. Dijkstra
Right. This is why I think VP9 actually could win and become the new standard (replacing H.264).
H.265 and VP9 seem like they are definitely in the same ballpark on quality. And H.265 is heavily encumbered with patents; you have to pay royalties, and you never know what the royalties might cost in five years. VP9, on the other hand, is simply free: no royalties, no restrictions on what you may do with the video.
Even if VP9 takes a lot more CPU time to encode, and even if H.265 is slightly better than VP9, not having to pay royalties (not even having to keep track of what you do with the video!) is such a huge benefit. It seems like a no-brainer.
And Google will be making sure that all the Android phones at least will have good hardware support for VP9 decoding. VP8 never had a chance against H.264 because the hardware support wasn't there, and large companies were content to pay the capped fees as you noted.
All that's left is possible legal FUD around VP9, but even that seems pretty cut-and-dried to me. MPEG-LA tried for something like a year to find patents to put into their patent pool to extract royalties from VP8, and in the end Google gave them a one-time payment of (to Google) a relatively small amount of money. Thanks to that one-time payment we know MPEG-LA won't ever come after anyone for using a VP8-derived codec, and I have no reason to think anyone would be able to prevail in court if they try it.
Given all of the above, it seems to me that VP9 is the obvious choice for the new video standard, and I kind of wonder why anyone is still interested in using H.265 and paying the royalties.
lf(1): it's like ls(1) but sorts filenames by extension, tersely