HDR heatmaps and torch mode UHDs

[fkid]

Indiana Jones and the Last Crusade (1989)

HDR BL vs. DoVi FEL HEATmaps/GAMUTmaps

HDR10 PLOT


Cheers!

[Fjodor2000]

Originally Posted by dorian

Same thing. The problem is PGS subtitles are gamma (SDR) encoded images that being ST 2084 (HDR) decoded by the display. Code values that ordinarily map to a reasonable light level in SDR get mapped to ridiculous light levels in HDR. As less than 1% of the frame exceeds 1000 nits, the subtitles will almost certainly hit display max too.

[Show spoiler]

It's the anti-aliasing on the text that clips and produces the 10,000 nit reading:

[Show spoiler]

The body of the text has code value 235 (in 8-bit). If we assign gamma a nominal peak luminance of 100 nits, we can see the crazy disparity in light output.
Column 1: code value. Column 2: gamma light output. Column 3: ST 2084 light output.

[Show spoiler]

Thanks a lot for your clear analysis and explanation. Much appreciated.

From your post, the body of the text has code value 235 (in 8-bit). From the luminance table, that means 82.20 nits in SDR, but 4773 nits in HDR.

Now if one would like to correct the subtitles to use a code value suitable for HDR, what should the code value 235 be replaced with on 4K BDs? And does it differ depending on whether the disc has been mastered using 4,000 or 10,000 nits as max brightness?

[dorian]

Originally Posted by Macatouille

Given that I'm new to this part, I'm open to suggestions for making them more accurate. Here's my complete screenshot command for reference:

ffmpeg -i "C:\Folder\file.mkv" -vf fps=1/30,zscale=t=linear:npl=100,format=gbrpf32le,zscale=p=bt709,tonemap=tonemap=hable:desat=0,zscale=t=bt709:m=bt709:r=tv,format=yuv420p C:\Folder\Heatmaps\FILE%04d.png

That looks fine. Hable's tonemap is as good as any. The only "true" way to represent HDR is if the content in the image doesn't exceed 100 nits because we can tonemap from ST 2084 to sRGB without clipping any code values. But most HDR shots do contain elements in the frame that exceed 100 nits which get clipped with a pure tonemap. The way to preserve them is to roll off the shoulder of the tone curve like Hable's tonemap does at the expense of decreased contrast. It's a necessary trade-off.

Here's a visual example comparing Hable's tonemap to a pure ST 2084 to sRGB tonemap

[dorian]

Originally Posted by Fjodor2000

Now if one would like to correct the subtitles to use a code value suitable for HDR, what should the code value 235 be replaced with on 4K BDs? And does it differ depending on whether the disc has been mastered using 4,000 or 10,000 nits as max brightness?

The mastering luminance makes no difference. I think there are ways to alter the brightness of PGS subtitles using CLI but it's quite involved. It would be a lot better if labels requested slightly darker PGS subtitles for HDR presentations. The solution is literally that simple: make the white subtitles grey/greyer.

[Fjodor2000]

Originally Posted by dorian

The mastering luminance makes no difference. I think there are ways to alter the brightness of PGS subtitles using CLI but it's quite involved. It would be a lot better if labels requested slightly darker PGS subtitles for HDR presentations. The solution is literally that simple: make the white subtitles grey/greyer.

Agreed. I was mainly looking for an instruction to whomever produces the 4K BDs. I.e. what code value would be suitable for them to use for the subtitles on 4K BDs? What value should they replace the code value 235 with?

I'm considering contacting Kino and some other production companies who have the same problem with blindingly bright subtitles on some of their 4K BDs. So it would be appreciated if clear instructions could be given to them about how to fix this problem on their 4K BD releases going forward.

[matbezlima]

Could someone here do The Nightmare Before Christmas 4K?

[Macatouille]

Tenet (2020)

Mastering display luminance              : min: 0.0050 cd/m2, max: 4000 cd/m2
Maximum Content Light Level              : 349 cd/m2
Maximum Frame-Average Light Level        : 86 cd/m2

Heatmaps (Album)

[Show spoiler]

Gamut Visualizations (Album)

[Show spoiler]

Tonemapped Screenshots (Album)

[Show spoiler]

HDR10 Plot

[Macatouille]

MaxCLL is just 406 nits and the MaxFALL 339. Doesn't try too far from the SDR luminance standards but does have pretty regular flashes. My brightest shots:



WCG gets used pretty regularly with the Red/Blue Room being the clear showcase:

[BobSimms]

Request for The Exorcist, if you please.

[matbezlima]

Originally Posted by matbezlima

Could someone here do The Nightmare Before Christmas 4K?

I would love to know, for example, the HDR heatmap and color gamut of the scene in the film's opening when Jack is on fire after having swallowed fire.

[Macatouille]

Originally Posted by matbezlima

Could someone here do The Nightmare Before Christmas 4K?

Originally Posted by BobSimms

Request for The Exorcist, if you please.

I plan on getting both of these but have no ETA as of now.

[dorian]

The Exorcist (1973)

MaxCLL plot:

[Show spoiler]

MaxFALL plot:

[Show spoiler]

The following shots are representative of the MLL & FALL peaks, not the averages.

MLL peaks:

[Show spoiler]

FALL peaks:

[Show spoiler]

Gamut Visualisation:

Keep in mind that colour transforms aren't absolutely perfect so seeing coordinates slightly out of gamut isn't a clear indication of a larger gamut. Another thing is that human perception of colourfulness and saturation is tied to light. Gamut plots show chromaticity coordinates independent of light. Screenshots can show highly chromatic colours but if the scene is dark it obviously won't look very colourful.

[Show spoiler]

[matbezlima]

Even without taking WCG into account, HDR will still have a major impact of colours. We naturally perceive some colors as brighter than others even when their intensity is the same. And increase in brightness also increases the perceive saturation if I'm not mistaken. So, HDR to SDR caps will often be very misleading and innacurate about the colors.

The fact that our perception of colors doesn't depend just on the wavelengths, but also the intensity if I'm not mistaken, means that talking about a color gamut is far from the whole story. The whole picture also would have to take the color volume into account. It's a color SPACE after all. A TV with good color gamut, but bad color volume, will often look muted and dull when displaying HDR content. It will often look bad. And the market is full of such TVs, these average TVs that can only decodify HDR signal, but are crap at displaying it, making HDR often look worse than SDR.

[Macatouille]

Blow Out (1981)

Mastering display luminance              : min: 0.0001 cd/m2, max: 1000 cd/m2
Maximum Content Light Level              : 1000 cd/m2
Maximum Frame-Average Light Level        : 95 cd/m2

Heatmaps (Album)

[Show spoiler]

Gamut Visualizations (Album)

[Show spoiler]

Tonemapped Screenshots (Album)

[Show spoiler]

HDR10 Plot

[Macatouille]

Plot shows a MaxCLL of 669 nits but I was able to capture 716 myself (just by grabbing a shot every 30 seconds):



Also saw several others north of 640. MaxFALL is just 67 nits (average of ~12) so the grade keeps the intentionally dark look in check. WCG usage seems decent here but not overly dramatic:

[dorian]

Originally Posted by matbezlima

Even without taking WCG into account, HDR will still have a major impact of colours. We naturally perceive some colors as brighter than others even when their intensity is the same. And increase in brightness also increases the perceive saturation if I'm not mistaken. So, HDR to SDR caps will often be very misleading and innacurate about the colors.

Right. I've said that HDR->SDR screenshots are only truly accurate with respect to colour and contrast up to 100 nits with a pure 1D tonemap. That's the reason tone mapping algorithms contain both luminance mapping and added saturation to *approximate* what you'll see in HDR.

The fact that our perception of colors doesn't depend just on the wavelengths, but also the intensity if I'm not mistaken, means that talking about a color gamut is far from the whole story. The whole picture also would have to take the color volume into account. It's a color SPACE after all. A TV with good color gamut, but bad color volume, will often look muted and dull when displaying HDR content. It will often look bad. And the market is full of such TVs, these average TVs that can only decodify HDR signal, but are crap at displaying it, making HDR often look worse than SDR.

I'm not sure I follow. The gamut we see on the chromaticity diagram is just a 2D projection of the 3D volume as viewed from top-down (x, y from xyY where Y is luminance). Is your point that TVs don't get bright enough to track the EOTF correctly?

[matbezlima]

Originally Posted by dorian

Right. I've said that HDR->SDR screenshots are only truly accurate with respect to colour and contrast up to 100 nits with a pure 1D tonemap. That's the reason tone mapping algorithms contain both luminance mapping and added saturation to *approximate* what you'll see in HDR.

Exactly, but even the best tone-mapping algorythms will often be wildly off. It's far from an exact science. There is, for example, the fact that we naturally perceive some color as brighter than others even when the intensity of light is actually the same (this is something very important that any big lights' event has to take into account). There are also other factors, which I will talk about in my reply below to the second quoted part of your comment.

Originally Posted by dorian

I'm not sure I follow. The gamut we see on the chromaticity diagram is just a 2D projection of the 3D volume as viewed from top-down (x, y from xyY where Y is luminance). Is your point that TVs don't get bright enough to track the EOTF correctly?

Yes, but also that the brightness of a given color also affects how saturated we perceive it to be, if I'm not mistaken (I might be wrong though). I think, but don't quote me on this, that a brighter color will look more saturated even if it's the same wavelength, or something like that. What I can say for certain is that a TV with decent color gamut, but poor color volume (how bright it can make each color), will really struggle with displaying any remotely bright color with the proper saturation, making the image look dull and muted. Color volume is tied to peak brightness and EOTF.

I'm talking this mostly from what RTINGS says regarding color gamut and color volume, they have one article for each.

https://www.rtings.com/tv/tests/pict...ci-p3-rec-2020

https://www.rtings.com/tv/tests/pict...3-and-rec-2020

But don't quote anything I'm saying here, I couldn't be further from an expert. I'm just trying my best to understand at least the essentials. HDR overall is hot confusing mess of a topic.

[Macatouille]

Gemini Man (2019)

Mastering display luminance              : min: 0.0010 cd/m2, max: 1000 cd/m2
Maximum Content Light Level              : 5180 cd/m2
Maximum Frame-Average Light Level        : 1577 cd/m2

Heatmaps (Album)

[Show spoiler]

Gamut Visualizations (Album)

[Show spoiler]

Tonemapped Screenshots (Album)

[Show spoiler]

HDR10 Plot

[Macatouille]

Looking at the thumbnails, and the plot, you can see how the look of the movie starts big and bright in the first half and then turns very dark in the second, while still delivering strong highlights. I generally think this looks great. A rare 60 FPS release with a Dolby Vision grade to boot. MaxCLL is 2043 nits, MaxFALL 1524.

Strongest highlights I grabbed:



WCG is used sparingly at first and then jumps about midway through:

[Fjodor2000]

Originally Posted by dorian

The mastering luminance makes no difference. I think there are ways to alter the brightness of PGS subtitles using CLI but it's quite involved. It would be a lot better if labels requested slightly darker PGS subtitles for HDR presentations. The solution is literally that simple: make the white subtitles grey/greyer.

Originally Posted by Fjodor2000

Agreed. I was mainly looking for an instruction to whomever produces the 4K BDs. I.e. what code value would be suitable for them to use for the subtitles on 4K BDs? What value should they replace the code value 235 with?

@dorian: Just to clarify:

If we assume code value 235 in SDR is suitable brightness for subtitles, what code value should be used to achieve the same brightness in HDR?

Code value 235 in SDR is 82.20 nits according to the table below. So what code value in HDR results in 82.20 nits?

PS. You only provided a screenshot of a section of the table below in your previous post, and no link to the full table, so I cannot look it up myself.



Found the table myself:

subtitles_sdr_hdr_nits.png

So the conclusion is that on the 4K BD the subtitles should have 8-bit brightness code value of 124 or 125, to result in 82 nits brightness in HDR, if correctly produced to match the 82 nits used on the BD. But since the 4K BD incorrectly has copied the 8-bit brightness code of 235 from the BD, the subtitles will have 4773 nits brightness in HDR on the 4K BD.