x265 and Dolby Vision

[Stacey Spears]

Originally Posted by benwaggoner

Yes, x265 absolutely supports encoding Dolby Vision, but only the single-layer Profiles 5, 8.1, and 8.2. So not the dual layer profiles. Profile 5 encodes the same quality at about 20% lower bitrate than dual layer, so that's what everyone is using today for offline playback unless they NEED the SDR backwards compatibility for some reason.

Blu-ray uses Profile 7, which requires the dual layer system. Profile 5 does not offer 20% savings over profile 7 if you are using MEL. It is just one vs. two files.

Does profile 5 support 12-bit? I have mostly worked with profile 7, so not sure if it does or not. With the full enhancement layer, you do get 12-bit support.

[Blue_MiSfit]

FWIU Profile 5 can reconstruct 12 bit RGB using the 10 bit IPT base layer and the RPU. They do want you to feed in a 12+ bit (ideally 16) when shaping the image into 10 bit IPT.

Ben was also saying before that you actually can do Profile 5 now on UHD-BD (not just 7)

[Stacey Spears]

Originally Posted by Blue_MiSfit

FWIU Profile 5 can reconstruct 12 bit RGB using the 10 bit IPT base layer and the RPU. They do want you to feed in a 12+ bit (ideally 16) when shaping the image into 10 bit IPT.

Ben was also saying before that you actually can do Profile 5 now on UHD-BD (not just 7)

Profile 5 cannot be used on BD, if Ben is suggestion it can, he would be mistaken.

The RPU does not contain any extra picture information, it just has the various Ls of information such as frame statistics and trim pass values. It is a pretty small file in binary form. Unfortunately BD does not support IPT either. I wish it did, my content shows noise in some super saturated red's due to non-constant luminance of YCbCr.

The output of the Dolby engine is always 12-bit, the LSBs are all 0 unless you encode 12-bit or you have a full enhancement layer to reconstruct the original 12-bit signal. IPT by itself does not offer any of that.

[ghostshadow]

ok that's it I managed to make an x265 encoded in dolby vision profile 8 from a UHD Dolby vision FEL Dual Track Dual Layer

[SeeMoreDigital]

Originally Posted by ghostshadow

ok that's it I managed to make an x265 encoded in dolby vision profile 8 from a UHD Dolby vision FEL Dual Track Dual Layer

Out of interest... What's the movie?

[ghostshadow]

Midway Fr

[Blue_MiSfit]

Originally Posted by Stacey Spears

The output of the Dolby engine is always 12-bit, the LSBs are all 0 unless you encode 12-bit or you have a full enhancement layer to reconstruct the original 12-bit signal. IPT by itself does not offer any of that.

Interesting. So what you're saying is that the output of profile 5 is effectively 10 bit, even though Dolby says the shaping of 12+ bit RGB input into 10 bit IPT lets them preserve more precision during playback?

[Stacey Spears]

Originally Posted by Blue_MiSfit

Interesting. So what you're saying is that the output of profile 5 is effectively 10 bit, even though Dolby says the shaping of 12+ bit RGB input into 10 bit IPT lets them preserve more precision during playback?

Yes. IPTs benefit is constant luminance. With YCbCr at high nit saturated colors, you get visible noise in the image. This all goes away in IPT. Netflix and others have developed techniques to try and reduce the noise, but not nearly as effective as IPT. After encoding, they modify the analyze the Y channel and alter it for cleaner results and then re-encode.

Have you ever noticed the dark transition between colors on a color bars test pattern? That is due to non-constant luminance of YCbCr. IPT does not suffer from this either.

PQ, in general, is at least two bits more efficient than gamma. 12-bit PQ is like 15-bit gamma.

[Blue_MiSfit]

Gotcha. Thanks for sharing.

So, the question becomes - what matters more? 10 bit PQ IPT / ICtCp etc to reduce high saturation noise, or the effective 12 bit YCbCr PQ delivered by DV Profile 7 with a full res enhancement layer?

[Stacey Spears]

Originally Posted by Blue_MiSfit

Gotcha. Thanks for sharing.

So, the question becomes - what matters more? 10 bit PQ IPT / ICtCp etc to reduce high saturation noise, or the effective 12 bit YCbCr PQ delivered by DV Profile 7 with a full res enhancement layer?

I don't know the answer. I included both FEL and MEL of the same clip and on today's consumer displays, hard to see much difference.

Streaming services are at a much lower bit rate than optical disc. That will make a difference.

I have the same rotating gradient pattern in both HDR10 and DV Profile 7 (MEL). On the C9, the DV has a smoother gradient. The DV is literally the HDR10 encode with dynamic metadata added, so the visual difference is the pipeline in the display. Dolby may simply have better dither, hard to say.

[Blue_MiSfit]

the visual difference is the pipeline in the display

I've absolutely found similar things. It seems Dolby does things "less wrong" inside the display than LG's default behavior with their HDR10 pipeline

[benwaggoner]

Originally Posted by Blue_MiSfit

I've absolutely found similar things. It seems Dolby does things "less wrong" inside the display than LG's default behavior with their HDR10 pipeline

DoVi does their own certification, so DoVi implementations tend to vary less than stock HDR-10 implementations.

LG's HDR-10 implementation got a lot better in the last few years compared to their early HDR models.

[Blue_MiSfit]

Agreed. The B6 was terrible, and the C9 is quite good!

[chainring]

FYI: Anyone seeing the weirdness of MediaInfo reporting that a freshly backed up title having been encoded with x265 and what looks to be the normal command line parameters from a normal encode, update to MediaInfo 20.09. It's also properly detecting the DoVI in titles ripped with MakeMKV 1.15.2 and later.

[Boulder]

Can anyone enlighten me - what is the proper procedure to create DV (and HDR10?) supported Matroska streams? I'm still wondering if it's possible to re-encode the normal video track to a lower bitrate. Chances are that Kodi will support DV in the future so it would mean that I could use those directly in my media player library.

[SeeMoreDigital]

Originally Posted by chainring

FYI: Anyone seeing the weirdness of MediaInfo reporting that a freshly backed up title having been encoded with x265 and what looks to be the normal command line parameters from a normal encode, update to MediaInfo 20.09. It's also properly detecting the DoVI in titles ripped with MakeMKV 1.15.2 and later.

In my tests it would appear that MediaInfo v20.09 can reliably detect Dolby Vision meta-data for 'single-layer' encoded HEVC streams muxed within MP4, MKV, M2TS and TS containers.

But it's currently a bit 'hit-and-mis' detecting Dolby Vision meta-data for 'dual-layer' encoded HEVC streams.

I've backed-up quite a few 4K UHD discs but the Dolby Vision meta-data is not detected when 'dual-layer' streams are muxed within their original M2TS container.

[benwaggoner]

Originally Posted by Blue_MiSfit

I've absolutely found similar things. It seems Dolby does things "less wrong" inside the display than LG's default behavior with their HDR10 pipeline

Yes, I'd say a core advantage of DoVi is that it's quality "floor" relative to the panel is higher than for HDR-10 implementations, due to Dolby's tuning and certification. Today's HDR-10 implementations from the major TV vendors are quite good, but there's a lot of variability outside of the top tier vendors and older models.

Plus dynamic metadata is a lot of help.

HDR10+ implementations, while not as proscribed as DoVi, are also very good from everything I've seen. The biggest challenge in tone mapping is how to adapt to potential variability, and static metadata isn't of much help. The best HDR-10 implementations do a lot of preanalysis of frames before display and statistical modeling. With dynamic metadata, there's no need to analyze and estimate, since the actual data is right there.

And while it's not possible to do much more than a frame or so of lookahead with dynamic metadata over HDMI (although even one frame is a lot of help), when doing playback of streams on a device, it's feasible to read the metadata many frames ahead for even smoother adaption, and more aggressive use of panel capabilities as less headroom is needed to reserve.

[jlpsvk]

ok... and now.. how can I get that RPU file for use with x265? maybe merge two BL.hevc with EL_RPU.hevc with yousesope's tool with "convert to 8.1"... this will create BL+RPU hevc stream, and then demux that hevc file and get BL.hevc and EL_RPU.hevc, but el_rpu.hevc should be only RPU file... which can be used for encoding with x265... did anyone tried that?

[SeeMoreDigital]

@jlpsvk,

You might want to checkout various discussions within this forums tsMuxer Open Source topic.

[Balling]

Originally Posted by Boulder

Can anyone enlighten me - what is the proper procedure to create DV (and HDR10?) supported Matroska streams? I'm still wondering if it's possible to re-encode the normal video track to a lower bitrate. Chances are that Kodi will support DV in the future so it would mean that I could use those directly in my media player library.

You use official matroksa tools. But I really like mkv less and less. No DTS (decoding timestamps), only millisecond precision (but there is a patch to fix time scale), ffmpeg think all mkvs are VFR, insane. I am so tired of ffmpeg's more and more insane behaviour.