Dolby Vision now possible through MP4 Mux.

[Rainer]

I did some tests whether DV is played on my TV some months ago.
That time I created mp4, ts & m2ts test files which worked.
I used that time ffmpeg and mp4muxer.

Later I realised that MakeMKV can also create an mkv with DV information.
Today I tried to use MakeMKV the first time to create the mkv. Aftwerwards I remuxed the mkv to m2ts using tsMuxer.
But my LG does not want to play that file at all.
What did I do wrong? Cannot I used the DV-mkv which MakeMKV created?

Code: Select all

Video
ID                                       : 4117 (0x1015)
Menu ID                                  : 1 (0x1)
Format                                   : HEVC
Format/Info                              : High Efficiency Video Coding
Format profile                           : Main 10@L5.1@High
HDR format                               : Dolby Vision, Version 1.0, dvhe.07.06, BL+EL+RPU, Blu-ray compatible / SMPTE ST 2086, HDR10 compatible
Codec ID                                 : 36
Duration                                 : 1 h 54 min
Width                                    : 3 840 pixels
Height                                   : 2 160 pixels
Display aspect ratio                     : 16:9
Frame rate                               : 23.976 (24000/1001) FPS
Color space                              : YUV
Chroma subsampling                       : 4:2:0 (Type 2)
Bit depth                                : 10 bits
Language                                 : English
Color range                              : Limited
Color primaries                          : BT.2020
Transfer characteristics                 : PQ
Matrix coefficients                      : BT.2020 non-constant
Mastering display color primaries        : Display P3
Mastering display luminance              : min: 0.0050 cd/m2, max: 1000 cd/m2
Maximum Content Light Level              : 600 cd/m2
Maximum Frame-Average Light Level        : 487 cd/m2

Edit:
Or which tool I can use to remux the mkv (created by MakeMKV) easily to m2ts (or mp4) and preserve the DV data?

[quietvoid]

Btw, since you are the DoVi guru (and we know a lot about hdr10), I don't really understand what the min (max) limunance setting does in reality. I don't see the point. Do you know anything about this? Thanks

I'm not sure what min/max you're talking about, in the RPU metadata?
It's usually the measured values for a frame/scene, so that's the dynamic part..

Also, setting the source min and max PQ in the RPU to proper values for limited has an effect.
Now I see down to 78 with 62/3696 (as spec says for profile 7), and 74 with 7/3079

Somehow 7/3079 is also being used in UHD BD.

From the data I can collect with the level 6 metadata (mastering display)..
Max 1000, min 1-10 should be 7/3079.
Max 4000, min 50 should be 62/3696.

Not sure about 10 000 nits but apparently Stacey Spears' sample would be correct by that logic.
Not sure why the image isn't though, so there might be a benefit of using different source values.

Maybe Dolby is crushing blacks a lot more for 10 000 nits mastered?

Here are samples to try:
7/3079: https://mega.nz/file/scFxEQob#UUiBMJBSV ... 1fwGJZ6URo
62/3696: https://mega.nz/file/YINjjYxK#IqOGwxla5 ... HTgern5QIs

[lexyz]

How does it look on your TVs?

Black levels seems mostly fine. Upclose to screen I can notice 64 slightly flashing, but not from viewing distance

But shield colors...

https://slow.pics/c/iHbxWsK2

[RESET_9999]

Unfortunately, it does not, it produces hue shift.
But the near black glowing (if you have it) can be eliminated with custom 1dlut!

Yep, I took a closer look and you are right. damn
I'll try your solution eventhough I'm a newbie for all the calibration and lut file import stuff. I paid someone to calibrate my TV but it is something that I want to learn. (bought an i1 display pro recently)

thank you

Personally if it was me I would just rip the iso and then play it back on a navidia shield tv as it has Dolby vision support and also Dolby atmos support

well, it is a good DV mediaplayer if you don't care about red push issue and the lack of FEL decoding support. HDR10 is probably better/more accurate than DV on the Shield.

In my case the bluray source has higher framecount than webdl dv source.
At the same frame , dv = 19641 and bluray = 19664
Will the procedure be same ?

Code: Select all

{
	"duplicate": [
		{
			"source": 0,
			"offset": 0,
			"length": 23
		}
	]
}

if you have a newer LG (C9 or newer i think), you might have to convert profile 7 to profile 8 (at least for the FEL movies)
check out this tool

But shield colors...

https://slow.pics/c/iHbxWsK2

oh wow, WTF... this doesnt happen on my shield + plex + C8
still crushed at 16 though

Maybe Dolby is crushing blacks a lot more for 10 000 nits mastered?

Here are samples to try:
7/3079: https://mega.nz/file/scFxEQob#UUiBMJBSV ... 1fwGJZ6URo
62/3696: https://mega.nz/file/YINjjYxK#IqOGwxla5 ... HTgern5QIs

oh wow... it is the first time I can see 66 flashing with a DV black pattern. Usually can only see around 72 flashing
even the p5 pattern which is mastered at 4000nits is crushed: i cant see below 16.

[quietvoid]

I'm actually seeing 72 flashing with L2 metadata target for 600 nits, source min/max 7,3079 and L6 metadata MDL 1000,1.
That was a generated RPU.

So with MDL 1000 metadata, LG player shows down to 68.
Firestick to 72.

So maybe this black level pattern is not really useful, thanks Dolby.

I believe that the ideal would be that, in case the amount of frames of the video and the RPU does not match, the last frame of the RPU is copied repeatedly the number of times necessary to obtain the same frame length as the video.
And in case the RPU contains more frames than the video, when the last frame of the video is processed with the RPU, skip the next frames of the RPU.

In fact, I think the tool should behave like this by default, without using that '--ignore-framecount' argument or any other argument.
But, when this happens, the tool should show a warning message about it so that the user is aware of it (just in case he did it accidentally).

Regards!

This has been implemented in v0.4.0

That's interesting what you talk about: can I add fake DoVi metadata (e.g. 1000 MDL, 1000 MaxCLL, 300 FALL) into a HDR10 stream, if so how?
(I want to create a small DoVi sample for FireStick 4k from a HDR10 sample to see whether it triggers tv-led or player-led with LG.)

You can now generate RPU metadata in v0.4.0, see documentation: https://github.com/quietvoid/dovi_tool/ ... nerator.md

You can find it here: https://github.com/quietvoid/dovi_tool/ ... /tag/0.4.0

[Rainer]

if you have a newer LG (C9 or newer i think), you might have to convert profile 7 to profile 8 (at least for the FEL movies)
check out this tool

Code: Select all

>dovi_tool info --input "test.mkv"
Parsing RPU file...
thread 'main' panicked at 'Input file probably too large', src\dovi\mod.rs:82:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

Where to set this?
Which profile does MakeMKV use?

Edit:
These tools do not seem to work when I have already a 1-layer-movie.
At least I did not manage it.

[RESET_9999]

I'm actually seeing 72 flashing with L2 metadata target for 600 nits, source min/max 7,3079 and L6 metadata MDL 1000,1.
That was a generated RPU.

So with MDL 1000 metadata, LG player shows down to 68.
Firestick to 72.

So maybe this black level pattern is not really useful, thanks Dolby.

Nah 66 is clearly visible using the C8 internal player with the 7/3079 sample but I have the perfect pitch black room with light absorbent material on walls, ceiling, and floor ( black velvet) and I'm young with a 20/20 vision.
I'm not sure what that means though, I would like to know what Stacey thinks about that.

This has been implemented in v0.4.0

wow this is great.. thank you works great... so much easier than before!!!!



also, I noticed that sometimes the maxcll/fall of the rpu are set to 0 while the hdr10 stream have metadatas. Do you recommend changing the metatadas of the rpu ?
BTW, I watched two p8 injected DV movies this weekend. Not a single issue. works just perfect. really amazing work man!!!

Edit:
These tools do not seem to work when I have already a 1-layer-movie.
At least I did not manage it.

the tools work. you have to demux the hevc stream first.

[quietvoid]

also, I noticed that sometimes the maxcll/fall of the rpu are set to 0 while the hdr10 stream have metadatas. Do you recommend changing the metatadas of the rpu ?

L6 metadata is optional as far as I know, but maybe.
It's not currently possible in the editor though but is something I wanted to add.

I just tried the S&M benchmark with generated Dolby Vision and it looks the exact same as the TV-led vs player-led comparison.
That's probably because the trim passes aren't present, so saturation is wrong.

Otherwise the tonemapping is a lot better than the default HDR10 on the C8 for 10000 nits..

[RESET_9999]

I just tried the S&M benchmark with generated Dolby Vision and it looks the exact same as the TV-led vs player-led comparison.
That's probably because the trim passes aren't present, so saturation is wrong.

Otherwise the tonemapping is a lot better than the default HDR10 on the C8 for 10000 nits..

that's right, I just tested this too. Once the rpu is modified(7/3079), DV becomes blown-out like(but not as much as) HDR10 @10k nits and i can see the same color difference as in the vincent video
So the LLDV bug is caused by the player not sending the correct trim pass when it's decoding DV? So the brighter the HDR grade is, the more the LLDV bug will be visible.

[quietvoid]

that's right, I just tested this too. Once the rpu is modified(7/3079), DV becomes blown-out like(but not as much) HDR10 @10k nits and i can see the same color difference as in the vincent video
So the LLDV bug is caused by the player not sending the correct trim pass when it's decoding DV? So the brighter the HDR grade is, the more the LLDV bug will be visible.

The bug is definitely the same thing, because creative adjustments aren't being applied by the player. I think Stacey mentioned that the L2 metadata was ignored, and that's what causes the image to be worse.
Obviously the brighter the grade, then L2 metadata has stronger adjustments depending on target display nits.

However that's all creative intent, so it would be complicated to add "proper" adjustments for non-DoVi HDR10 movies.

[RESET_9999]

yep, it all makes sense now.

the eagle beak still remains relatively yellow though while in LLDV with original fel, it turns greenish.

[chros]

I have the perfect pitch black room with light absorbent material on walls, ceiling, and floor ( black velvet)

That's a hardcore setup, congrats!

Btw, since you are the DoVi guru (and we know a lot about hdr10), I don't really understand what the min (max) limunance setting does in reality. I don't see the point. Do you know anything about this? Thanks

I'm not sure what min/max you're talking about, in the RPU metadata?
It's usually the measured values for a frame/scene, so that's the dynamic part..

Also, setting the source min and max PQ in the RPU to proper values for limited has an effect.
Now I see down to 78 with 62/3696 (as spec says for profile 7), and 74 with 7/3079

Somehow 7/3079 is also being used in UHD BD.

From the data I can collect with the level 6 metadata (mastering display)..
Max 1000, min 1-10 should be 7/3079.
Max 4000, min 50 should be 62/3696.

oh wow... it is the first time I can see 66 flashing with a DV black pattern. Usually can only see around 72 flashing

I'm actually seeing 72 flashing with L2 metadata target for 600 nits, source min/max 7,3079 and L6 metadata MDL 1000,1. That was a generated RPU.
So with MDL 1000 metadata, LG player shows down to 68. Firestick to 72.
So maybe this black level pattern is not really useful, thanks Dolby.

Nah 66 is clearly visible using the C8 internal player with the 7/3079 sample ... I'm not sure what that means though, I would like to know what Stacey thinks about that.

That's a really interesting finding!
I meant the measured values for a frame/scene in the RPU, but I thought there's a min value as well, but there isn't, only maxCLL and maxFall per frame.
But you now brought up the "Source min and max PQ", what does it mean exactly? And what are those "7/3079" and "62/3696" numbers corresponding to? Which metric? (I bet it's not nits.)
And what is the L1 min_pq/max_pq/avg_pq?

Here all the 3 sources (internal Plex, Oppo, Firestick4k) produces different results with those files
Firestick4k is the best with the 3079 sample (as you said), here I can see 74 flashing, but it cuts everything under it, no matter which picture preset I select.
The brightest source out of all is the internal Plex, 2nd is Oppo, then Firestick at the last place (Maybe HDMI signal cuts black at certain point as well, not sure.)

This has been implemented in v0.4.0
You can now generate RPU metadata in v0.4.0, see documentation

wow this is great.. thank you works great... so much easier than before!!!!
...
BTW, I watched two p8 injected DV movies this weekend. Not a single issue. works just perfect. really amazing work man!!!

Indeed, You have created an amazing, useful tool, thank You!
E.g. now we can easily create DoVi test patterns out of our HDR10 ones
Here's the previously posted DP p8.1 DoVi banding sample (05-2160p_23fps_dovi_p81-dp_s02e03_banding.ts) with generated RPU (I measured the maxFALL with madvr's madmeasure):

Code: Select all

{
    "length": 1000,
    "target_nits": 1000,
    "level6": {
        "max_display_mastering_luminance": 1000,
        "min_display_mastering_luminance": 1,
        "max_content_light_level": 1000,
        "max_frame_average_light_level": 67
    }
}

What is the "target_nits" for? (for L2 target_max_pq?) Is it in nits? If so, can we use any value here? (e.g. 2435)
Initially I used 600 (as in your sample), but the result didn't "pop", after that I did it with 1000 and it's much better this way.
So, if you also agree, then this value needs to be changed in your sample as well.

I just tried the S&M benchmark with generated Dolby Vision and it looks the exact same as the TV-led vs player-led comparison.
That's probably because the trim passes aren't present, so saturation is wrong.

So the LLDV bug is caused by the player not sending the correct trim pass when it's decoding DV?

I think Stacey mentioned that the L2 metadata was ignored, and that's what causes the image to be worse.

I guess you meant the 10.000 nits SM version.
Nice finding, again! But what is a trim pass in DoVi world and what is it used for? (I saw it in the RPU as well.)

[Manixx2020beyound]

I have the perfect pitch black room with light absorbent material on walls, ceiling, and floor ( black velvet)

That's a hardcore setup, congrats!

Btw, since you are the DoVi guru (and we know a lot about hdr10), I don't really understand what the min (max) limunance setting does in reality. I don't see the point. Do you know anything about this? Thanks

I'm not sure what min/max you're talking about, in the RPU metadata?
It's usually the measured values for a frame/scene, so that's the dynamic part..

Also, setting the source min and max PQ in the RPU to proper values for limited has an effect.
Now I see down to 78 with 62/3696 (as spec says for profile 7), and 74 with 7/3079

Somehow 7/3079 is also being used in UHD BD.

From the data I can collect with the level 6 metadata (mastering display)..
Max 1000, min 1-10 should be 7/3079.
Max 4000, min 50 should be 62/3696.

oh wow... it is the first time I can see 66 flashing with a DV black pattern. Usually can only see around 72 flashing

I'm actually seeing 72 flashing with L2 metadata target for 600 nits, source min/max 7,3079 and L6 metadata MDL 1000,1. That was a generated RPU.
So with MDL 1000 metadata, LG player shows down to 68. Firestick to 72.
So maybe this black level pattern is not really useful, thanks Dolby.

Nah 66 is clearly visible using the C8 internal player with the 7/3079 sample ... I'm not sure what that means though, I would like to know what Stacey thinks about that.

That's a really interesting finding!
I meant the measured values for a frame/scene in the RPU, but I thought there's a min value as well, but there isn't, only maxCLL and maxFall per frame.
But you now brought up the "Source min and max PQ", what does it mean exactly? And what are those "7/3079" and "62/3696" numbers corresponding to? Which metric? (I bet it's not nits.)
And what is the L1 min_pq/max_pq/avg_pq?

Here all the 3 sources (internal Plex, Oppo, Firestick4k) produces different results with those files
Firestick4k is the best with the 3079 sample (as you said), here I can see 74 flashing, but it cuts everything under it, no matter which picture preset I select.
The brightest source out of all is the internal Plex, 2nd is Oppo, then Firestick at the last place (Maybe HDMI signal cuts black at certain point as well, not sure.)

This has been implemented in v0.4.0
You can now generate RPU metadata in v0.4.0, see documentation

wow this is great.. thank you works great... so much easier than before!!!!
...
BTW, I watched two p8 injected DV movies this weekend. Not a single issue. works just perfect. really amazing work man!!!

Indeed, You have created an amazing, useful tool, thank You!
E.g. now we can easily create DoVi test patterns out of our HDR10 ones
Here's the previously posted DP p8.1 DoVi banding sample (05-2160p_23fps_dovi_p81-dp_s02e03_banding.ts) with generated RPU (I measured the maxFALL with madvr's madmeasure):

Code: Select all

{
    "length": 1000,
    "target_nits": 1000,
    "level6": {
        "max_display_mastering_luminance": 1000,
        "min_display_mastering_luminance": 1,
        "max_content_light_level": 1000,
        "max_frame_average_light_level": 67
    }
}

What is the "target_nits" for? (for L2 target_max_pq?) Is it in nits? If so, can we use any value here? (e.g. 2435)
Initially I used 600 (as in your sample), but the result didn't "pop", after that I did it with 1000 and it's much better this way.
So, if you also agree, then this value needs to be changed in your sample as well.

I just tried the S&M benchmark with generated Dolby Vision and it looks the exact same as the TV-led vs player-led comparison.
That's probably because the trim passes aren't present, so saturation is wrong.

So the LLDV bug is caused by the player not sending the correct trim pass when it's decoding DV?

I think Stacey mentioned that the L2 metadata was ignored, and that's what causes the image to be worse.

I guess you meant the 10.000 nits SM version.
Nice finding, again! But what is a trim pass in DoVi world and what is it used for? (I saw it in the RPU as well.)

Trim pass process.
https://learning.dolby.com/hc/en-us/art ... Trim-Pass-

[quietvoid]

That's a really interesting finding!
I meant the measured values for a frame/scene in the RPU, but I thought there's a min value as well, but there isn't, only maxCLL and maxFall per frame.
But you now brought up the "Source min and max PQ", what does it mean exactly? And what are those "7/3079" and "62/3696" numbers corresponding to? Which metric? (I bet it's not nits.)
And what is the L1 min_pq/max_pq/avg_pq?

Source min and max PQ are PQ code values in 12 bit, converting to nits should be simple, it's just a division by 4095 and converting from PQ to nits.
I'm not exactly sure what they specify, but I think it might have to do with the mastering display, since they seem correlated to the L6 metadata.

L1 metadata are measurements done per frame/scene (depends) by analysis. It's supposed to help with the tonemapping.

What is the "target_nits" for? (for L2 target_max_pq?) Is it in nits? If so, can we use any value here? (e.g. 2435)
Initially I used 600 (as in your sample), but the result didn't "pop", after that I did it with 1000 and it's much better this way.
So, if you also agree, then this value needs to be changed in your sample as well.

Target nits is for L2 target_max_pq, yes. It is in nits.
Normally it is used for creative adjustments per trim ("different displays"), because the grading is done at multiple brightness to map properly across the whole range while keeping creative intent.
I didn't think a single L2 target would make any difference, but maybe. Currently everything is left default, so no adjustments at all.

From what I noticed, highlights are tonemapped OK but I'm not sure if FALL is lower. Saturation might be different for higher nits.
I didn't really compare properly with different targets.

I don't really expect generated metadata to be very useful, since it's missing creative intent and measurements.
And since measurements often depend on denoised video (L4 metadata) and properly separated scenes, it would be quite complex to do.

[RESET_9999]

Hi guys im getting crazy with this, i dont even know if its possible or not.

I have a dolby vision single layer mp4, ripped from web, inside theres multiple audio tracks so i just want to have the dolby vision single layer with 1 track.

there is no EL in profile 5 DV from streaming.
demux, then :

Code: Select all

mp4muxer_64bits.exe --dv-profile 5 --input-file X:\video.hevc --input-file X:\audio.ec3 --output-file output.mp4

Hi, sorry for the noob question but how and what do i need to use to extract both layer from the dv web mp4 rip? for dovi tool i need the hevc already but how i go from the dv mp4 ripped from web and the hevc keeping both layers?
with mp4demux i only get 1 hevc that doest contain the dolby layer...

you have to extract and convert the RPU to profile 8(step #2). Once you know the frame difference with the bluray remux, you make a json and you can inject the rpu from the web-dl.