Dolby Vision now possible through MP4 Mux.

[quietvoid]

The interesting part is that how grey it is depends on the metadata! E.g. from the previously linked Dolby manual (thanks for thjis for you RESET and Mannix):
"... Blu-ray players ... Due to this implementation on these devices, any positive lift applied to Dolby Vision content during the “trim pass,” will raise the black levels in the letterbox/blanking areas of the image and can become distracting to the end user. Dolby recommends a maximum positive lift value of 0.025 while doing the trims on letterboxed content during the Dolby Vision content creation process."

This is not the same problem, the lift affects letterbox bars that are already present in the content.
But in our case there are grey bars when the content does not have letterbox bars at all, because they were cropped out.

[chros]

This is not the same problem, the lift affects letterbox bars that are already present in the content.
But in our case there are grey bars when the content does not have letterbox bars at all, because they were cropped out.

Good point if it's cropped already, I talked about the active image area metadata that simulates cropping.

[RESET_9999]

just generated a profile 08.06 from an HDR10 sample, plays fine with internal Plex app: 04-2160p_23fps_dovi_p81-l_s02e01_cyan_error.ts

To me, it looks like madVR is including the black bars in the measurement because all the min L1 values are 0 in your DV sample. Could be just a different scene cut though.

If I don't set the correct blanking in resolve, I also get 0 everywhere. If set correctly, then I get my first min 0 at exactly 14sec ( frame 343)
otherwise, the max and avg numbers are pretty close. I did not check for a visual difference, only checked the metadata.


I generated metadata using Davinci Resolve Dolby Vision algorithm(same algo studios use). Here's how the L1 metadata compares to madVR for this Netflix clip.





madvr rpu frame 1:

Code: Select all

{
        "Level1": {
          "block_info": {
            "ext_block_length": 5,
            "ext_block_level": 1,
            "remaining": [
              0,
              0,
              0,
              0
            ]
          },
          "min_pq": 0,
          "max_pq": 3157,
          "avg_pq": 1075
        }
      },
      {
        "Level2": {
          "block_info": {
            "ext_block_length": 11,
            "ext_block_level": 2,
            "remaining": [
              0,
              0,
              0
            ]
          },
          "target_max_pq": 3079,
          "trim_slope": 2048,
          "trim_offset": 2048,
          "trim_power": 2048,
          "trim_chroma_weight": 2048,
          "trim_saturation_gain": 2048,
          "ms_weight": 2048
        }
      },
      {
        "Level5": {
          "block_info": {
            "ext_block_length": 7,
            "ext_block_level": 5,
            "remaining": [
              0,
              0,
              0,
              0
            ]
          },
          "active_area_left_offset": 0,
          "active_area_right_offset": 0,
          "active_area_top_offset": 0,
          "active_area_bottom_offset": 0
        }
      },
      {
        "Level6": {
          "block_info": {
            "ext_block_length": 8,
            "ext_block_level": 6,
            "remaining": []
          },
          "max_display_mastering_luminance": 1000,
          "min_display_mastering_luminance": 1,
          "max_content_light_level": 1000,
          "max_frame_average_light_level": 58

resolve rpu frame 1:

Code: Select all

 {
        "Level1": {
          "block_info": {
            "ext_block_length": 5,
            "ext_block_level": 1,
            "remaining": [
              0,
              0,
              0,
              0
            ]
          },
          "min_pq": 12,
          "max_pq": 3298,
          "avg_pq": 2048
        }
      },
      {
        "Level2": {
          "block_info": {
            "ext_block_length": 11,
            "ext_block_level": 2,
            "remaining": [
              0,
              0,
              0
            ]
          },
          "target_max_pq": 2081,
          "trim_slope": 1533,
          "trim_offset": 2007,
          "trim_power": 1177,
          "trim_chroma_weight": 2048,
          "trim_saturation_gain": 2048,
          "ms_weight": 2048
        }
      },
      {
        "Level2": {
          "block_info": {
            "ext_block_length": 11,
            "ext_block_level": 2,
            "remaining": [
              0,
              0,
              0
            ]
          },
          "target_max_pq": 2851,
          "trim_slope": 2058,
          "trim_offset": 2030,
          "trim_power": 2053,
          "trim_chroma_weight": 2048,
          "trim_saturation_gain": 2048,
          "ms_weight": 2048
        }
      },
      {
        "Level2": {
          "block_info": {
            "ext_block_length": 11,
            "ext_block_level": 2,
            "remaining": [
              0,
              0,
              0
            ]
          },
          "target_max_pq": 3079,
          "trim_slope": 2101,
          "trim_offset": 2035,
          "trim_power": 2219,
          "trim_chroma_weight": 2048,
          "trim_saturation_gain": 2048,
          "ms_weight": 2048
        }
      },
      {
        "Level3": {
          "block_info": {
            "ext_block_length": 2,
            "ext_block_level": 3,
            "remaining": []
          },
          "min_pq_offset": 2077,
          "max_pq_offset": 1432,
          "avg_pq_offset": 2048
        }
      },
      {
        "Level5": {
          "block_info": {
            "ext_block_length": 7,
            "ext_block_level": 5,
            "remaining": [
              0,
              0,
              0,
              0
            ]
          },
          "active_area_left_offset": 0,
          "active_area_right_offset": 0,
          "active_area_top_offset": 120,
          "active_area_bottom_offset": 120
        }
      },
      {
        "Level6": {
          "block_info": {
            "ext_block_length": 8,
            "ext_block_level": 6,
            "remaining": []
          },
          "max_display_mastering_luminance": 1000,
          "min_display_mastering_luminance": 1,
          "max_content_light_level": 1680,
          "max_frame_average_light_level": 58

[chros]

just generated a profile 08.06 from an HDR10 sample, plays fine with internal Plex app: 04-2160p_23fps_dovi_p81-l_s02e01_cyan_error.ts

To me, it looks like madVR is including the black bars in the measurement because all the min L1 values are 0 in your DV sample. Could be just a different scene cut though.

That's 0 because madvr doesn't produce any info about min_pq per frame, jast as hdr10+ (as quietvoid and manixx suggested)

I generated metadata using Davinci Resolve Dolby Vision algorithm(same algo studios use). Here's how the L1 metadata compares to madVR for this Netflix clip.

It's Amazon hdr10+. Thanks for the test, measurement results are pretty close indeed (apart from the first avg value, where it's double), really interesting:
- Resolve L6 maxCLL is set to 1680, but it measure more than 3000 nits? what's the unit of the values on the image, nits or PQ?
- - interestingly that's the measurement summary from madvr: MaxCLL 100%: 1680, 99.9%: 1545, MaxFALL: 58, AvgFALL: 9, AvgFMLL: 524 nits
- Resolve set some L2 trim properties (that obviously dovi_tool doesn't by default), why is that? Did you set it or is it automatic?
- Resolve has 3 blocks of L2 metadata, why is that? (target_max_pq: 2081, 2851, 3079; dovi_tool has just 1: 3079) Which one is used during playback?

Min pq is mostly set to 0 on hdr10plus ... n true min pq value dv dynamic meta it varies more often

If I don't set the correct blanking in resolve, I also get 0 everywhere. If set correctly, then I get my first min 0 at exactly 14sec ( frame 343)
otherwise, the max and avg numbers are pretty close. I did not check for a visual difference, only checked the metadata.

Thanks, my question is what the effect of min_pq value is in reality.
@quietvoid, can we overwrite an L2 value in RPU easily, if so how? (e.g. min_pq to 12) Or we have to export it into json and manually replace them and then reinjecting it?

[quietvoid]

- Resolve has 3 blocks of L2 metadata, why is that? (target_max_pq: 2081, 2851, 3079; dovi_tool has just 1: 3079) Which one is used during playback?

One block for every target display peak brightness, so that creatives can adjust to their liking for the whole range of brightness.
In playback, you'd have to figure out what the TV says it can output, this might be the same peak as in the Dolby Vision configuration files used for calibration.
This is why I set multiple blocks to the same values in my L2 test sample.

@quietvoid, can we overwrite an L2 value in RPU easily, if so how? (e.g. min_pq to 12) Or we have to export it into json and manually replace them and then reinjecting it?

You're asking for L2 but min_pq is L1 metadata.
Technically it could be added to the editor but I don't really see a good use case for it..

[Manixx2020beyound]

D9036A05-52EB-42CD-83D5-2E3CB37AD41D.png (803.88 KiB) Viewed 13677 times

Be interesting to find out what mad Vr considers as the active area.
Did anyone try a movie without black bars to see if the min values changed?

[Manixx2020beyound]

- Resolve has 3 blocks of L2 metadata, why is that? (target_max_pq: 2081, 2851, 3079; dovi_tool has just 1: 3079) Which one is used during playback?

One block for every target display peak brightness, so that creatives can adjust to their liking for the whole range of brightness.
In playback, you'd have to figure out what the TV says it can output, this might be the same peak as in the Dolby Vision configuration files used for calibration.
This is why I set multiple blocks to the same values in my L2 test sample.

@quietvoid, can we overwrite an L2 value in RPU easily, if so how? (e.g. min_pq to 12) Or we have to export it into json and manually replace them and then reinjecting it?

You're asking for L2 but min_pq is L1 metadata.
Technically it could be added to the editor but I don't really see a good use case for it..

I don’t think anyone needs l2/l8 trims if the creators intent is already within the source
U can select to bypass trim pass for l2 which won’t produce any l2 meta in resolve.
Plus isn’t l2 for mapping trims to 709
L8 trims maps to 1000/600nits dv which we have no acess to without a license which is blacked out in resolve

[RESET_9999]

Thanks, my question is what the effect of min_pq value is in reality.

I don't really know but Dolby clearly says that min 0 for all the scenes is incorrect.
When I get home, I'll do the analysis with the HDR10plus algorithm and see if I get a different min pq for that clip.

It's Amazon hdr10+. Thanks for the test, measurement results are pretty close indeed (apart from the first avg value, where it's double), really interesting:
1- Resolve L6 maxCLL is set to 1680, but it measure more than 3000 nits? what's the unit of the values on the image, nits or PQ?
- - interestingly that's the measurement summary from madvr: MaxCLL 100%: 1680, 99.9%: 1545, MaxFALL: 58, AvgFALL: 9, AvgFMLL: 524 nits
2- Resolve set some L2 trim properties (that obviously dovi_tool doesn't by default), why is that? Did you set it or is it automatic?
3- Resolve has 3 blocks of L2 metadata, why is that? (target_max_pq: 2081, 2851, 3079; dovi_tool has just 1: 3079) Which one is used during playback?

Thanks, my question is what the effect of min_pq value is in reality.

1- the L6 metadata is edited manually, I just took the numbers from your post: viewtopic.php?p=113519#p113519
2- this is the resolve default trim passes adjustments for each target display. You need a license to modify them in resolve but everything else is available and free to use.
3- As Quietvoid said.

For the foreseeable future, the current consumer display landscape encompasses a wide variety of
differently performing televisions and projectors that are guaranteed to improve year over year. This
means that mastering for today’s displays may render content less vibrant than content that emerges
five years from now. This can be especially vexing for narrative content that will have a long lifespan
on streaming services as new generations of viewers discover them. While one way of solving this
would be to re-grade your program many times at a variety of nit levels to create deliverables suitable
to a range of display capabilities, that’s an enormous amount of work.
Dolby Vision offers a shortcut by using sophisticated algorithms to derive automatically analyzed
metadata that intelligently guides how an image graded at one nit level (say 4000 nits) can be
adjusted to be perceptually similar to viewers watching a 1000 nit display. Highlights and saturation
that are too bright for a particular display will be adjusted to provide as close to the same experience
without clipping or flattening image detail.
Furthermore, this automatic analysis can be manually trimmed by a colorist to account for the artistic
intentions of the authors of a program, in cases where the automatic analysis doesn’t do exactly what’s
wanted. This combination of auto-analysis and manual trimming is key to how Dolby Vision streamlines
the process of mastering programs to accommodate backward compatibility with SDR displays, as well
as the varying peak luminance capabilities of different makes and models of HDR consumer displays,

both now and in the future. You’re only required to make a 100 nit trim pass to guide the HDR
program’s conversion all the way down to SDR, and the Dolby Vision system can use that information
to guide how intermediate presentations (such as at 700 or 1200 nits) should be adjusted. You can
even do multiple trim passes at specific nit levels, such as a 100 nit pass and a 1000 nit pass, to give
the Dolby Vision system more information to accurately guide intermediate presentations on different
displays. Additionally, you don’t have to trim every clip. If the analysis is good, you can skip those clips
and only trim clips that need it. The overall system has been created to make it as efficient as possible
for colorists to ensure that the widest variety of viewers see the image as it’s meant to be seen.
This, in a nutshell, is the advantage of the Dolby Vision system. You can grade a program on a more
future-proofed 4000 nit display, and use auto-analysis plus one or two manual trim passes to make the
program backward compatible with SDR televisions, and capable of intelligently scaling the HDR
highlights to provide the best representation of the mastered image for whatever peak luminance and
color volume a particular television is capable of. All of this is guided by decisions made by the colorist
during the grade.
At the time of this writing, all seven major Hollywood studios are mastering in Dolby Vision for cinema.
Studios that have pledged support to master content in Dolby Vision for home distribution include
Universal, Warner Brothers, Sony Pictures, and MGM. Content providers that have agreed to distribute
streaming Dolby Vision content include Netflix, Vudu, and Amazon. If you want to watch Dolby Vision
content on television at home, consumer television manufacturers LG, TCL, Vizio, HiSense, Sony,
Toshiba, and Bang & Olfusen have all shipped models with Dolby Vision supports

[quietvoid]

Be interesting to find out what mad Vr considers as the active area.
Did anyone try a movie without black bars to see if the min values changed?

Current behavior is hardcoded to min_pq=0 for both HDR10+ and madVR generate.
madVR only provides the luminance histogram, which includes the letterbox bars. So it's complicated to figure out the actual minimum of the active area.

I don't really know but Dolby clearly says that min 0 for all the scenes is incorrect.
When I get home, I'll do the analysis with the HDR10plus algorithm and see if I get a different min pq for that clip.

At best the min would be the same as source_min_pq for all the shots, instead of 0.
I don't think this is the case of most content, though. I've seen 0 used.

It's just to have accurate L1, but I don't think the value 0 is wrong itself.

[chros]

In playback, you'd have to figure out what the TV says it can output, this might be the same peak as in the Dolby Vision configuration files used for calibration.
This is why I set multiple blocks to the same values in my L2 test sample.

So, do we know what happens when only 1 L2 block is present? I mean do we have to specify all of them or it will use the only one for all?
Also, what nit number do these correspond to? 2081, 2851, 3079

You're asking for L2 but min_pq is L1 metadata.
Technically it could be added to the editor but I don't really see a good use case for it..

My bad, sorry, L1. No worries, editing json is fine.

Be interesting to find out what mad Vr considers as the active area. Did anyone try a movie without black bars to see if the min values changed?

It's not that, but the binary measurement file doesn't have any data about this.

I don’t think anyone needs l2/l8 trims if the creators intent is already within the source
U can select to bypass trim pass for l2 which won’t produce any l2 meta in resolve.

Good point.

Plus isn’t l2 for mapping trims to 709
L8 trims maps to 1000/600nits dv which we have no acess to without a license which is blacked out in resolve

According to the manual L8 is for secondary trim controls only in v4.0 (and not backward compatible with v2.9)

I don't really know but Dolby clearly says that min 0 for all the scenes is incorrect.
When I get home, I'll do the analysis with the HDR10plus algorithm and see if I get a different min pq for that clip.

Yes, that's why I asked it. But this is a good point that then we have to use active area as well (otherwise it will be incorrect this time )

It's Amazon hdr10+. Thanks for the test, measurement results are pretty close indeed (apart from the first avg value, where it's double), really interesting:
1- Resolve L6 maxCLL is set to 1680, but it measure more than 3000 nits? what's the unit of the values on the image, nits or PQ?
- - interestingly that's the measurement summary from madvr: MaxCLL 100%: 1680, 99.9%: 1545, MaxFALL: 58, AvgFALL: 9, AvgFMLL: 524 nits
2- Resolve set some L2 trim properties (that obviously dovi_tool doesn't by default), why is that? Did you set it or is it automatic?
3- Resolve has 3 blocks of L2 metadata, why is that? (target_max_pq: 2081, 2851, 3079; dovi_tool has just 1: 3079) Which one is used during playback?

Thanks, my question is what the effect of min_pq value is in reality.

1- the L6 metadata is edited manually, I just took the numbers from your post: viewtopic.php?p=113519#p113519
2- this is the resolve default trim passes adjustments for each target display. You need a license to modify them in resolve but everything else is available and free to use.
3- As Quietvoid said.

1- Oh, I see
2 - interesting, thanks

[quietvoid]

So, do we know what happens when only 1 L2 block is present? I mean do we have to specify all of them or it will use the only one for all?
Also, what nit number do these correspond to? 2081, 2851, 3079

I haven't tested, but I'd assume Dolby interpolates between targets on playback.
Or at least uses the trims for the closest target.

The value is PQ code, so 2081/4095 is the PQ code for 100 nits.
2851 = 600, 3079 = 1000, etc.

Some movies only have trims for 100 nits, but most include at least 600 as well.

My bad, sorry, L1. No worries, editing json is fine.

I don't even think the exported JSON can be made back into a binary RPU..

[PlatypusW]

Infuse on Apple TV just got updated to support profile 5 in mkv. Don’t know if that’s a reasonable alternative to shield for p5 mkv files now?
Lack of lossless audio is still an issue but most profile 5 files won’t have lossless audio anyway.

[NiCE77]

Infuse on Apple TV just got updated to support profile 5 in mkv. Don’t know if that’s a reasonable alternative to shield for p5 mkv files now?
Lack of lossless audio is still an issue but most profile 5 files won’t have lossless audio anyway.

All P5 files are webrip, so there are not too many movies uploaded in this version, the P8 support would be much better.

[jesper1994]

Hi. Sorry to ask something already covered in this thread, but I haven't been able to figure out the latest/best info.

Back in Dec 2020, I started using scripts from this thread to convert Dolby Vision UHD discs (Game of Thrones) to DV MP4, but I never finished the process.

What's the best option currently to convert these discs to a DV file that will play on AppleTV with LG C8 — or using LG C8 built-in player?

GoT episodes are in FEL.
On C7, C8, C9, the best option is to make DV dual layers mp4 files (dvhe.07.06) and to play them directly with the LG built-in player.
For each episode, extract BL, EL and audio files with tsmuxer, then mux the mp4 file with mp4muxer.
The result is almost 1.8 TB for 98 files (mp4 + srt files)

Do we still need to add the audio with mp4box or can we use mp4muxer?

[daddy]

Hi. Sorry to ask something already covered in this thread, but I haven't been able to figure out the latest/best info.

Back in Dec 2020, I started using scripts from this thread to convert Dolby Vision UHD discs (Game of Thrones) to DV MP4, but I never finished the process.

What's the best option currently to convert these discs to a DV file that will play on AppleTV with LG C8 — or using LG C8 built-in player?

GoT episodes are in FEL.
On C7, C8, C9, the best option is to make DV dual layers mp4 files (dvhe.07.06) and to play them directly with the LG built-in player.
For each episode, extract BL, EL and audio files with tsmuxer, then mux the mp4 file with mp4muxer.
The result is almost 1.8 TB for 98 files (mp4 + srt files)

Do we still need to add the audio with mp4box or can we use mp4muxer?

When the audio is in ac3 or eac3 you can mux it directly with the BL and EL with mp4muxer.
When the audio is in dts or dts-hd you have to mux BL and EL with mp4muxer, and then add the audio to the mp4 with mp4box