HEVC and H264 bitrate/quality equivalency

[philvideofilm]

Hi all,

Recently updated my comp and now I want to jump on the HEVC bandwagon. I have done my fair share of research but there is one tidbit of info I can't seem to find. In general terms what are the equivalencies, in terms of encoding settings, between something I would have encoded in H264 but now does so in HEVC, in order of course, to achieve the same visual fidelity?

I understand that quality is dependent on the content but I am not looking for a specific setting, just a general ballpark estimate that would guide me in my tests ahead.

Thanks

[jagabo]

Generally, HEVC can get the same quality as AVC at somewhere between 50 and 75 percent the bitrate. But a lot depends on the encoder and settings used.

[philvideofilm]

Yes this is what I have seen floating around but it still somehow feels vague for me, I know I am stupid.

Any chance that you, or others, know of a test that was posted here or somewhere else and that it is the same files but encoded in both format and show the same quality? So I could easily download them, analyze them and see for myself?

Thanks

[jagabo]

But that would only tell you about those particular videos, those particular encoders, and those particular settings. There are lots of such comparisons like that here. Most comparing hardware encoders to software encoders. Search for things like x264, x265, CUDA, NvEnc, QSVEnc, etc.

[philvideofilm]

I fully understand that but it would be of tremendous help to me if I had access to such a test, because right now I feel like my arms are flapping the waters and my efforts/tests are not going anywhere.

Thanks

[jagabo]

This is the most systematic of the tests I've seen: http://compression.ru/video/codec_comparison/index_en.html

[Hoser Rob]

... a lot depends on the encoder and settings used.

And I think that counts for whoever's doing the encode as well as the software. I've seen terrible quality HEVC encoded video. I could have done better in the same bit rate with Xvid, and I'm no encoding guru.

[_Al_]

Detail seems to be gone using HEVC, using x265 for example (comparing to x264), so I'd say it is meant for higher resolutions so far, where there is enough pixels in the first place , like 4k and that bandwith savings can get huge also. Especially web streaming, downloads. Not as such for example encoding SD resolution, personally I cannot imagine encoding SD resolution to HEVC, what for.

But x265 is still in development. If it ain't broken do not fix it. Maybe someone would fix libvpx for VP9, maybe Eve (for VP9) will be free or some versions of it, maybe VP10/AV1 will surprise. Not like there is the end of the World for new codecs. No need to be in a hurry and encode into those latest codecs or worse to even re-encode.

To test HEVC just encode some sample with CRF using x264 and x265 and first you find apropriate CRF's getting the same volumes, then compare quality, that's a starter. Then you start to increase quantizer with x265 to find out how high you can go to have visually the same result. Use 10bit for HEVC, not sure why to bother with 8bit. Then you compare volumes. You might be surprised. This is the only reason to use HEVC - justified volume difference. And it gets interesting only with higher resolutions as I mentioned. Not sure why even to bother with fullHD (1920x1080). But you have people that encode to very little bitrates, starving encoders, there is even people posting here that would shock you with how low bitrates they are testing. HEVC might perform better regarding this. It really depends how you encode. There is no general answer.

[jagabo]

Also note that the MSU tests use single pass ABR mode. That's really only useful for realtime streaming. It doesn't give as good results as 2-pass VBR or constant quality encoding at similar bitrates. And they measure quality with YUV-SSIM, which doesn't necessarily relate well to "visual" quality.

Obviously, given enough biterate, all modern codecs can produce video indistinguishable (at normal viewing speed and distance) from the source. The issue is how much bitrate is needed and what types of artifacts become visible (and how obvious they are) at lower bitrates.

[philvideofilm]

Allright, after extensive testing I am happy with the quality but I can't seem to be able to bring down the file size. Take a look at the Mediainfo for both files :



My test, which something I shot is 1080p 60fps, 1:50, and at a QBR of 18 and the encoding speed set to slow speed it still gives me a file size of 125mb.

Compare that to the tv episode of 57 minute with a filesize of 277mb, and for that size the quality is astonishing.

Granted the episode is 720 and not 1080 but it shouldn't make that much a of a differencr, right?

Do I use the wrong settings? Am I using the right encoder (handbrake)?

Thanks

[hello_hello]

Also note that the MSU tests use single pass ABR mode. That's really only useful for realtime streaming. It doesn't give as good results as 2-pass VBR or constant quality encoding at similar bitrates. And they measure quality with YUV-SSIM, which doesn't necessarily relate well to "visual" quality.

I came across this recently. It's a few years old but it seems to me the effort they put into making the comparison between the reference h265 encoder and x264 "fair", crippled x264's compression ability, then they declared h265 produced the same quality (according to PSNR) with something like a 40% bitrate reduction.

The gop size was set to eight frames, and I-Frames were forced once every second. Each test encode only consisted of 150 frames. They mention adjusting the QP settings, but I'm not clear as to what was done there.
I thought the reference to x264's 2 pass encoding was interesting, because they almost make it sound like 2 pass encoding is a recent x264 innovation.

Performance Comparison of H.265/MPEG-HEVC, VP9, and H.264/MPEG-AVC Encoders (pdf)

[jagabo]

Granted the episode is 720 and not 1080 but it shouldn't make that much a of a differencr, right?

Wrong. 1920x1080 is over twice as many pixels as 1280x720 so it will take more bitrate. And your 1080p source is 60 fps, not 24 fps. So that will require more bitrate too. And your handheld, noisy, camcorder video probably won't compress as well as professionally shot film.

[philvideofilm]

Actually it is professionally video as I am a director/videographer with 12 years of experience and so the source is not shaky at all


My original questions remain, in the screenshots does any of you see anything wrong with my settings? Google pretty much always points to Handbrake when I search for the "best" HEVC encoder, but is it really or are there other encoders out there who do the job better?

Why, for example, the episode is set to 680KB bitrate and yet it looks phenomenal but for my file the encoder, set at 18QBR, felt the need to allocate 9mb of bitrate and it pretty much looks the same?

From reading guides I am under the impression that QBR maximizes quality by minimizing bitrates so that it is a better option over setting an average bitrate?

I understand that if a file is already compressed, thinking that recompressing it will yield a better result is a pipe dream. And yet, the episode is from a web release, and thus already compressed, and the result is amazing. How does that work.

Thanks

[hello_hello]

I know nothing about x265 settings but the file sizes are somewhat relative.

Take a pristine lossless source and encode it with mpeg2 at a high bitrate.
Take that same source again and encode it with mpeg2 at a low bitrate.

Now re-encode the two mpeg2 encodes with x264 or x265. You wouldn't expect the same bitrate reduction each time, because the bitrates of the sources are so different. At best, the two x265 encodes will result in the same bitrate for a given quality setting, but just as likely, the re-encode of the low bitrate mpeg2 encode will result in a slightly higher bitrate than re-encoding the first, because at a low bitrate the mpeg2 encoder probably intoduced encoding artefacts, and when you re-encode that source, the encoder has to try to re-encode the artefacts too.

Even if the bitrates are the same after re-encoding the two mpeg2 files, obviously one encode will have achieved a greater bitrate reduction than the other, so when re-encoding the first mpeg2 video you can marvel at how efficiently x265 compresses, yet after re-encoding the second one you might complain about the lack of bitrate reduction.

When you're using a quality based encoding method and the same encoder settings each time, the bitrate is determined by how hard the source is to recompress (as well as resolution and frame rate etc).

[philvideofilm]

Good stuff Hello_Hello, thanks!

[criggs]

personally I cannot imagine encoding SD resolution to HEVC, what for.

Oh, I can answer that.

I have a brother who has a wretched Internet connection. He literally cannot play anything back that streams at a higher bit rate than 900k. Yes, 900kilobuts. Not kilobytes, kilobits (it's an old DSL connection; his building is still waiting for fiber).

Which means that most HD video is out of his reach, and the only stuff he can watch is half-SD or worse material.

So I've taken to converting some of the web videos he wants to watch to 265, and posting those for him privately on my web site. But I found that creating an HD video that runs at 900 k, even in h.265, still looks like crap. On the other hand, if I compromise, and give the 265 video a resolution of 720 x 406, it looks acceptable; not great but acceptable.

So there's an example where 265 SD can be helpful.