HDR Discussion Thread

[Penton-Man]

Originally Posted by mrtickleuk

Is there any technical reason why HDR10+ metadata cannot have a manual pass?

If the HDR10+ metadata is in the form of SEI messages in the video stream (apologies if my terminology is not correct there), what evidence does anyone have how that metadata got there?.....

To generate HDR10plus metadata the user needs to load the HDR master encoded with the PQ curve and add to the timeline. The statistics generation is implemented as a metadata encoder, called: HDR10 Plus. After adding the encoder, the user needs to render the file, which will be a human readable JSON file containing the mathematical coefficients defining the tone map curve for each scene. By default the encoder automatically detects cuts between scenes based on the light level statistics.



Encoder to generate the HDR10plus metadata



The metadata in the generated file targets a specific brightness level, defined in nits in the TargetLevel parameter of the encoder. The maximum brightness of the source master should be set in the MasteringLevel setting. To QC or visualize the result of the dynamic remap, a HDR10plus node is added to the node page.

Then the user can import the generated JSON file through the Timeline menu by clicking on Cut Detection with HDR10plus JSON. This operation adds the cuts defined in the JSON file as well as populates the HDR10plus node with the appropriate metadata. As soon as the metadata is loaded into the nodes, the remap is processed. To validate, park it on a frame with bright pixels exceeding the nit level defined in the TargetLevel of the metadata, and toggle Bypass on the HDR10plus node.



HDR10plus node

[Penton-Man]

Originally Posted by JohnAV

Thats a bit generalizing something that can help notice visual differences on a static frame versus a continuous visual presentation. I don’t doubt a complex algorithm could work to make the image quality better, but we have two other infamous examples such as interpolation where you artificially add non real information to make the image appear higher quality, increases details/colors presentation as well as the long debated usage of Darbee where the enhanced image appears to have increased sharpness and contrast. Neither in the end compared to higher resolution content, just a algorithm solution to artificial image enhancement.

Good counterpoints.

[Penton-Man]

Originally Posted by PaulGo

Are you trying to hammer home a point?

Peter, Paul and Mary - If I Had A Hammer (1963 performance) - YouTube

[gkolb]

Originally Posted by Penton-Man

To generate HDR10plus metadata the user needs to load the HDR master encoded with the PQ curve and add to the timeline. The statistics generation is implemented as a metadata encoder, called: HDR10 Plus. After adding the encoder, the user needs to render the file, which will be a human readable JSON file containing the mathematical coefficients defining the tone map curve for each scene. By default the encoder automatically detects cuts between scenes based on the light level statistics.

[Show spoiler]
Encoder to generate the HDR10plus metadata

The metadata in the generated file targets a specific brightness level, defined in nits in the TargetLevel parameter of the encoder. The maximum brightness of the source master should be set in the MasteringLevel setting. To QC or visualize the result of the dynamic remap, a HDR10plus node is added to the node page.

Then the user can import the generated JSON file through the Timeline menu by clicking on Cut Detection with HDR10plus JSON. This operation adds the cuts defined in the JSON file as well as populates the HDR10plus node with the appropriate metadata. As soon as the metadata is loaded into the nodes, the remap is processed. To validate, park it on a frame with bright pixels exceeding the nit level defined in the TargetLevel of the metadata, and toggle Bypass on the HDR10plus node.

[Show spoiler]
HDR10plus node

Nice explanation! (Which I surely don't understand.)

A question though, is any of this dynamic metadata oriented towards a type of display, say LCD, and not OLED? Will the metadata recognize a lower capable nit OLED and perform differently than if the display is a 1500 nit LCD?

Besides DoVi's newly revealed dependence on calibration and look up tables, where does Samsung HDR10+ leave nits on the table, so to speak, and just let the display tone map and roll off highlights? Did I miss some other weakness?

[Geoff D]

I could be wrong (and I'm not answering for Penton here) but metadata is basically display-format agnostic despite what kind of display it might've been graded on. It doesn't 'do' anything by itself because the onus is and has always been on the display's processing to read the incoming data and to map that information appropriately.

[Penton-Man]

https://forum.blu-ray.com/showthread...y#post14562554

Follow-up to ^.
I was graciously bestowed one ^ (and a Team jersey, at that, not available for purchase in their online store).

Perhaps I’ll post a pic of it sometime in the future for all you closet mountain biker maniacs, but don’t all collectively hold your breaths as last time I was awarded something at CES (one of dem fancy crystal trophies in '08), it took me dang near over 7 months to post a pic after I announced it -
https://forum.blu-ray.com/showthread...esk#post508663
https://forum.blu-ray.com/showthread...re#post1103264

[Penton-Man]

Originally Posted by gkolb

Nice explanation! (Which I surely don't understand.)

Bottom line - Of what is being done manually with HDR10+….isn’t much.

Originally Posted by gkolb

A question though, is any of this dynamic metadata oriented towards a type of display, say LCD, and not OLED?

Nope, the type of display doesn’t enter into it. At its core, the key contributing component to the HDR10+ approach is the maxRGB parameter and how Samsung utilizes it, specifically -

average_maxrgb[ w ] specifies the average of linearized maxRGB values in the w-th processing window in the scene. The value of average_maxrgb[ w ] shall be in the range of 0 to 1, inclusive and in multiples of 0.00001.

num_distribution_maxrgb_percentiles[ w ] indicates the number of linearized maxRGB values at given percentiles in the w-th processing window in the scene. The maximum value of num_distribution_maxrgb_percentiles[ w ] shall be 15.

distribution_maxrgb_percentages[ w ][ i ] specifies an integer percentage value corresponding to the i-th percentile linearized RGB value in the w-th processing window in the scene. The value of distribution_maxrgb_percentages[ w ][ i ] shall be in the range of 0 to100, inclusive.

distribution_maxrgb_percentiles[ w ][ i ] specifies the linearized maxRGB value at the i-th percentile in the w-th processing window in the scene. The value of distribution_maxrgb_percentiles[ w ][ i ] shall be in the range of 0 to 1, inclusive and in multiples of 0.00001

(https://www.sra.samsung.com/assets/U...94-40-v1.1.pdf)

[Penton-Man]

Originally Posted by Geoff D

I could be wrong (and I'm not answering for Penton here) but metadata is basically display-format agnostic despite what kind of display it might've been graded on. It doesn't 'do' anything by itself because the onus is and has always been on the display's processing to read the incoming data and to map that information appropriately.

You’re not wrong.
Next time I’ll scroll down the page before I answer, cause if I’d had, I’d of been out of here by now and emptying the garbage which my wife is now scolding me for/to do.

[BrownianMotion]

Originally Posted by Penton-Man

Bottom line - Of what is being done manually with HDR10+….isn’t much.

Nope, the type of display doesn’t enter into it. At its core, the key contributing component to the HDR10+ approach is the maxRGB parameter and how Samsung utilizes it, specifically -

average_maxrgb[ w ] specifies the average of linearized maxRGB values in the w-th processing window in the scene. The value of average_maxrgb[ w ] shall be in the range of 0 to 1, inclusive and in multiples of 0.00001.

num_distribution_maxrgb_percentiles[ w ] indicates the number of linearized maxRGB values at given percentiles in the w-th processing window in the scene. The maximum value of num_distribution_maxrgb_percentiles[ w ] shall be 15.

distribution_maxrgb_percentages[ w ][ i ] specifies an integer percentage value corresponding to the i-th percentile linearized RGB value in the w-th processing window in the scene. The value of distribution_maxrgb_percentages[ w ][ i ] shall be in the range of 0 to100, inclusive.

distribution_maxrgb_percentiles[ w ][ i ] specifies the linearized maxRGB value at the i-th percentile in the w-th processing window in the scene. The value of distribution_maxrgb_percentiles[ w ][ i ] shall be in the range of 0 to 1, inclusive and in multiples of 0.00001

(https://www.sra.samsung.com/assets/U...94-40-v1.1.pdf)

In that pdf you posted, it says:

The metadata are content-dependent and can vary scene-by-scene or frame-by-frame

I remember reading that HDR10+ metadata varied only on a scene-by-scene basis, and not frame-by frame.

[PaulGo]

Originally Posted by BrownianMotion

In that pdf you posted, it says:



I remember reading that HDR10+ metadata varied only on a scene-by-scene basis, and not frame-by frame.

https://www.flatpanelshd.com/news.ph...&id=1463138030

https://www.cnet.com/news/what-is-hdr10/

[BrownianMotion]

Originally Posted by PaulGo

https://www.flatpanelshd.com/news.ph...&id=1463138030

That's based on a Samsung press release.

Samsung promotional and technical material on the subject suggests that HDR10+ metadata can vary on a frame-by-frame basis, but other experts (I think Stacey Spears may be who I'm thinking of) have claimed that it can only vary on a scene-by-scene basis.

[DanBa]

Latest Yoeri Geutskens' diagram:
https://twitter.com/UHD4k/status/952689567462805504

[mrtickleuk]

Originally Posted by Penton-Man

Bottom line - Of what is being done manually with HDR10+….isn’t much.

Many thanks for your awesome detailed replies!

Would it be fair to say that in theory, the metadata could be edited by hand(*), but it looks as if there isn't a nice UI front-end for doing it at the moment alongside the video.

*if it's a JSON file

[Penton-Man]

Originally Posted by BrownianMotion

Stacey Spears may be

Where is Stacey?

I hope he didn’t fall into the Grand Prismatic Spring and is deep in work with Ian (http://www.imdb.com/name/nm1324569/?ref_=nv_sr_1) as March will be here before we know it.

[2themax]

Originally Posted by BrownianMotion

That's based on a Samsung press release.

Samsung promotional and technical material on the subject suggests that HDR10+ metadata can vary on a frame-by-frame basis, but other experts (I think Stacey Spears may be who I'm thinking of) have claimed that it can only vary on a scene-by-scene basis.

I can confirm it's scene-by-scene on disc. Maybe streaming is different but I doubt it. I'm sure it would violate a Dolby patent if it was frame based.

[Penton-Man]

Originally Posted by mrtickleuk

Many thanks for your awesome detailed replies!

Would it be fair to say that in theory, the metadata could be edited by hand....

You're welcome.
In theory, yes, I think that would be fair to say.

Some believe their solution further upstream could have been even better in terms of image quality if they’d increased the 2nd italicized parameter (num_distribution_maxrgb_percentiles) to greater than 15, but alas, perhaps more processor intensive.

[Penton-Man]

Originally Posted by DanBa

Latest Yoeri Geutskens' diagram:
https://twitter.com/UHD4k/status/952689567462805504

P.S.
How’s that Venn diagram work out for those viewers visually challenged with some types of color blindness?
Better? for the induced vertigo.

[Penton-Man]

Originally Posted by 2themax

I can confirm it's scene-by-scene on disc....

x2

Excerpted from a post from last September, pertinent part now bolded -

Originally Posted by Penton-Man

Although I am more of a Dolby Vision than HDR10+ proponent for specific reasons stated long ago (e.g. DV 's more elegant tone mapping algorithm - https://forum.blu-ray.com/showthread...4#post12007747 , 12-bit support, true frame-by-frame accuracy with no lag),

[BrownianMotion]

Samsung lying in their promotional and technical papers about HDR10+. Not sure how much of a difference scene-by-scene vs frame-by-frame makes in practice, but it's very interesting that they would lie about it.

[Geoff D]

Originally Posted by Penton-Man

P.S.
How’s that Venn diagram work out for those viewers visually challenged with some types of color blindness?
Better? for the induced vertigo.

Heh, I'm not saying this just to Hang with Mr Penton but I had a similar reaction to that diagram when I saw it above, I took one look and thought "nah, that's much too busy for me".