HDR heatmaps and torch mode UHDs

[nissling]

Originally Posted by Fjodor2000

Do you suspect that the German edition has passed the SDR source material through some SDR-to-HDR-omatic processing?

It appears to have been converted from BT.709 to BT.2100 with the color space conversion tool in Resolve or something similar, and probably have had some adjustments done to it aftwards. I really dislike its appearence due to its high peaks and averages, but I don't condemn SDR to HDR conversions or SDR like HDR grades automatically, as those can still be done in a respectful manner.

Ghost Dog for instance is mastered at 100 nits white level in HDR, and barely exceeds that level. You could essentially call it SDR and question the decision of even utilizing HDR10, but the final image itself is still very good and enjoyable. Had the same thing been done with Triangle of Sadness, I surely wouldn't care.

As to SDR to HDR conversions made by a TV, I've never investigated that. Since the SDR master of Triangle of Sadness that's present on the Criterion is already excellent, I'd highly recommend sticking to viewing that natively.

[Fjodor2000]

Originally Posted by nissling

I really dislike its appearence due to its high peaks and averages

Well there are plenty 4K HDR masters with high peaks and averages. I’m not talking about such masters that have been first scanned in SDR and then converted to HDR, but masters that have been scanned in HDR directly from the film negative (or recorded in HDR directly if digital).

Do you think it’s a problem with high peaks and averages in such cases too? Even if the film is depicting e.g bright daylight content and/or specular highlights? In that case what should we even use HDR for, if we’re only going to use brightness levels close to SDR anyway?

[nissling]

I think you overanalyze my posts completely.

Originally Posted by Fjodor2000

I’m not talking about such masters that have been first scanned in SDR and then converted to HDR, but masters that have been scanned in HDR directly from the film negative (or recorded in HDR directly if digital).

Films are never "scanned in HDR", just like digital cameras don't "record in HDR". Most scanners use either a linear or logarithmic response curve to properly maintain the dynamic range of the film. On negatives this is actually fairly easy thanks to Kodak Cineon Log, which makes it fully possible to properly emulate the tone curve of a negative in the digital domain and can be fully handled in just about any color grading system (Resolve, Nucoda, Baselight etc).

Originally Posted by Fjodor2000

Do you think it’s a problem with high peaks and averages in such cases too? Even if the film is depicting e.g bright daylight content and/or specular highlights?

I don't like to generelize. I've never said there is a problem with high peaks nor high averages, although much of the impact of HDR is clearly being lost if peaks and averages are high all the time. It's similar to the way loudness war works.

Out of all the thousands of films I've scanned and worked on, I've done maybe a handful of complete restorations in HDR (considering how small the market is we can only justify doing it on very select titles). I always use an existing reference to get an idea how it originally looked (most of the time an archived print as long as it's in a usable state) and aim to maintain a similar average luminance level and let the highlights get as bright as I need them to. Examples below from a film I restored last year (B/w 35mm negative, mid 50s).

Average is at 10 nits but peak is still kind of low. I decided to not make the lamp too bright as it would end up looking to distracting.

[Show spoiler]

This one is perhaps more clear. Averages is at around 10 nits which is perfectly fine but the peak is very obvious, both at its luminance level as well as how much of the image it takes up (this scene has a torture sequence where the guards use extremely bright lights on a prisoner).

[Show spoiler]

I prefer to do HDR grades that are reserved, where the peaks are clearly visible without overbrightening the entire image. Some studios seem to stick to this way of doing their grades, like Warner Brothers, while others are far more into torch grades (though it has gotten way better in the last four or five years).

Originally Posted by Fjodor2000

In that case what should we even use HDR for, if we’re only going to use brightness levels close to SDR anyway?

SDR is relying on power law EOTF, which is essentially only emulating the transfer curve of a CRT and works in a fully relative manner. HDR puts it aside completely and relies on nominal values. This is crucial for creative intent as the image can always be delivered with the proper light levels to the end user. Not to mention, you are never limited by the container itself and can make the image as bright as it needs to be without introducing any clipping. For a colorist, this plays a big role.

[Fjodor2000]

Originally Posted by nissling

Films are never "scanned in HDR", just like digital cameras don't "record in HDR". Most scanners use either a linear or logarithmic response curve to properly maintain the dynamic range of the film. On negatives this is actually fairly easy thanks to Kodak Cineon Log, which makes it fully possible to properly emulate the tone curve of a negative in the digital domain and can be fully handled in just about any color grading system (Resolve, Nucoda, Baselight etc).

My point is that when you do a 4K master, you can select whether to output it in SDR or HDR format. If selecting to use SDR instead of HDR, the max brightness level that can be represented is lower.

The film negative scanning is just a part of the process (and it's N/A for digital recording).

(And I know you are already familiar with all of this, I'm just clarifying the point I'm trying to make.)

Originally Posted by nissling

I don't like to generelize. I've never said there is a problem with high peaks nor high averages, although much of the impact of HDR is clearly being lost if peaks and averages are high all the time. It's similar to the way loudness war works.

Out of all the thousands of films I've scanned and worked on, I've done maybe a handful of complete restorations in HDR (considering how small the market is we can only justify doing it on very select titles). I always use an existing reference to get an idea how it originally looked (most of the time an archived print as long as it's in a usable state) and aim to maintain a similar average luminance level and let the highlights get as bright as I need them to. Examples below from a film I restored last year (B/w 35mm negative, mid 50s).

Average is at 10 nits but peak is still kind of low. I decided to not make the lamp too bright as it would end up looking to distracting.

[Show spoiler]

This one is perhaps more clear. Averages is at around 10 nits which is perfectly fine but the peak is very obvious, both at its luminance level as well as how much of the image it takes up (this scene has a torture sequence where the guards use extremely bright lights on a prisoner).

[Show spoiler]

I prefer to do HDR grades that are reserved, where the peaks are clearly visible without overbrightening the entire image. Some studios seem to stick to this way of doing their grades, like Warner Brothers, while others are far more into torch grades (though it has gotten way better in the last four or five years).



SDR is relying on power law EOTF, which is essentially only emulating the transfer curve of a CRT and works in a fully relative manner. HDR puts it aside completely and relies on nominal values. This is crucial for creative intent as the image can always be delivered with the proper light levels to the end user. Not to mention, you are never limited by the container itself and can make the image as bright as it needs to be without introducing any clipping. For a colorist, this plays a big role.

What it boils down to is whether you think SDR is sufficient to represent whatever video you intend to represent. Surely, SDR is nowhere close to being able to represent reality as we view it with our eyes at least. Not even HDR is, but it's closer.

You can of course argue that there is always the creator's intent. For example perhaps the director doesn't want it to look bright, not even the frame average brightness in any of the scenes, regardless if shot under very bright conditions. That's totally fine. But then we're no longer talking about representing reality as accurately as possible. It can even be e.g. black & white, and then no color is needed.

So in the case of Triangle of Sadness, let's assume the intent was to represent reality as close as possible. Would the max frame average, frame average, and/or peak brightness then be higher than the SDR edition, if HDR had been used throughout when recording and producing the 4K master? At least for parts of the movie? After all, several scenes are shot under bright daylight conditions and some e.g. on open sea. So there ought to be some scenes with quite high frame average brightness and specular highlight brightness.

If you think the max frame average, frame average, and/or peak brightness would be higher if HDR had been used, then I take it your complaint is more about how the SDR-to-HDR conversion has been applied in the German HDR edition? E.g. perhaps the peak brightness should be higher than the SDR edition, but the German HDR edition boosted it too much, or applied it in the wrong places?

[nissling]

Originally Posted by Fjodor2000

But then we're no longer talking about representing reality as accurately as possible.

Which we never discussed in the first place as the film was never intended to "represent reality". It's intended to represent Östlund's vision, which he has created along with his team.

Film doesn't look like reality. Never did, and probably never will.

Originally Posted by Fjodor2000

So in the case of Triangle of Sadness, let's assume the intent was to represent reality as close as possible. Would the max frame average, frame average, and/or peak brightness then be higher than the SDR edition, if HDR had been used throughout when recording and producing the 4K master? At least for parts of the movie? After all, several scenes are shot under bright daylight conditions and some e.g. on open sea. So there ought to be some scenes with quite high frame average brightness and specular highlight brightness.

Again this is a completely irrelevant question for the reasons I already stated. The intent is what Östlund had in mind, and that can practically speaking be "whatever". It may be brighter, it may be the same, it may be darker. I don't know.

Originally Posted by Fjodor2000

If you think the max frame average, frame average, and/or peak brightness would be higher if HDR had been used, then I take it your complaint is more about how the SDR-to-HDR conversion has been applied in the German HDR edition? E.g. perhaps the peak brightness should be higher than the SDR edition, but the German HDR edition boosted it too much, or applied it in the wrong places?

My complaints are the following: The German disc is essentially a cheap torch grade, done with a simple color space conversion tool more or less, that has been done without the director's approval or involvement. It does not only deviate from the director's intention, but also introduces new issues in the image that are not present on the Criterion.

If they would have gotten in touch with Östlund or his team (they are not that difficult to get hold of) and done the work by the book, then it would most likely have turned out very different all together. But then again, it's waaay cheaper this way so... I'm gonna go for doubt.

By the way, the torture scene that I mentioned looks like this in one sequence. It appears as really bright as the clip right before it has a way lower peak (less than 300) and reserved average. It gives an incredible impact.

[Show spoiler]

This one is kinda interesting too. Peak is at 1000 nits but... Good luck noticing it. The 99th percentile is certainly way below that. I'm including a tonemapped screenshot just for the sake of it. I think it looks great.

[Show spoiler]

[dorian]

Originally Posted by Fjodor2000

If selecting to use SDR instead of HDR, the max brightness level that can be represented is lower.

That isn't true. The maximum output is dictated solely by the display. The conflation with transfer functions and brightness is perhaps the worst thing to come from the HDR mess. Think about it. Why would a power function limit the maximum brightness compared to a different encoding function when they both operate in the same [0, 1] domain?

There has to be a clear demarcation when talking about HDR/SDR masters and the encoding. Grading monitors can reach the same peaks for both. The difference is in the authorial intent, which itself is odd tbf. Should music be mastered differently for speakers with greater output?

The only technical difference between consumer HDR/SDR releases is the transfer function used to encode the image. Contrary to marketing bs, the benefits of the PQ/ST2084 transfer function have little to do with brightness, and much to do with bit efficiency. It's literally in the title of Dolby's presentation of PQ back in 2012: 'Perceptual Signal Coding for More Efficient Usage of Bit Codes', claiming that a new transfer function is needed to avoid visible quantisation as the maximum output of displays increases lol.

The film negative scanning is just a part of the process (and it's N/A for digital recording).

Nissling went over this but it's worth reiterating that there isn't really such a thing as HDR scanning. Scanners that claim to do HDR are mostly feeding in to the hype. All they do is take two passes at different gains to preserve whatever clips in the first pass. The problem (not actually a problem) is the first pass is always sufficient. Film has, at most, 14 stops of usable latitude. People will claim film has more because the density asymptotes as opposed to a hard clip à la digital but that is not the case.

Digital sensors measure light linearly until the signal saturates. The signal itself is an analogue voltage that is fed to the ADC and stored as a linear code value or log encoded for bit efficiency. The job of the ADC is to represent the incoming linear signal with sufficient precision, where one stop of light conveniently corresponds to one bit of storage. 14-bit ADCs have existed for a long time so whatever latitude is available on film has already been captured.

What it boils down to is whether you think SDR is sufficient to represent whatever video you intend to represent. Surely, SDR is nowhere close to being able to represent reality as we view it with our eyes at least. Not even HDR is, but it's closer.

Why is representing reality in an absolute fashion the artistic goal? You will never escape the fact that you are viewing an image on a windowed screen. The human visual system is not absolute. We have a limited range that shifts up and down as we adapt the environment. Adaptation isn't instant either, which is why large, sudden changes in APL hurt our eyes. Even gradients within the same frame can sometimes be so steep that things look weird.

Originally Posted by nissling

HDR puts it aside completely and relies on nominal values. This is crucial for creative intent as the image can always be delivered with the proper light levels to the end user.

You're right that the absolute transfer function does help to preserve creative intent, but it comes with the major caveat of being in the same environment as it was authored. Something consumers are not privy to, nor do they always have the luxury of attaining. I wonder how people would feel about absolute sound where you had no control over the volume in a film in order to respect creative intent.

[nissling]

Originally Posted by dorian

You're right that the absolute transfer function does help to preserve creative intent, but it comes with the major caveat of being in the same environment as it was authored. Something consumers are not privy to, nor do they always have the luxury of attaining. I wonder how people would feel about absolute sound where you had no control over the volume in a film in order to respect creative intent.

You are absolutely right about this.

SDR grading should be done with both dark and brighter enviroments taken into account, as the luminance will be scaled to whichever brightness the display is outputting. With HDR, you only have to focus on making the image look accurate in a properly dark enviroment.

It makes things easier for the artist, but it demands more from the consumer.

EDIT: And again, big thanks for all your work with your work which makes it possible for me and other entusiasts to present these analyses.

[mrtickleuk]

Originally Posted by dorian

You're right that the absolute transfer function does help to preserve creative intent, but it comes with the major caveat of being in the same environment as it was authored. Something consumers are not privy to, nor do they always have the luxury of attaining. I wonder how people would feel about absolute sound where you had no control over the volume in a film in order to respect creative intent.

Thanks for the whole post. I really love that last point!

[Geoff D]

Originally Posted by nissling

You are absolutely right about this.

SDR grading should be done with both dark and brighter enviroments taken into account, as the luminance will be scaled to whichever brightness the display is outputting. With HDR, you only have to focus on making the image look accurate in a properly dark enviroment.

It makes things easier for the artist, but it demands more from the consumer.

EDIT: And again, big thanks for all your work with your work which makes it possible for me and other entusiasts to present these analyses.

As much as HDR is a 'dark room' format, it's not just the consumer (and their marketing-led expectations of what HDR is and does) who's having more demanded of them by an absolute transfer function, it's the TV manufacturers themselves who dropped the ball big time by not having some standardised method of HDR10 tone mapping. Instead they all did/do different things and on top of that there was/is the aggressive anti-static dimming features of certain display types which really didn't play nice with transfers that had very low APL. And they seemed to forget about projektors entirely such were the problems that people had with tone mapping there.

So the ironical thing about HDR providing so much more precision for the artiste is that the home implementation of it is the most wildly unstable and inconsistent rollout of such a tech that I've ever seen. It's a wonder that any two people get to see even vague resemblances when watching the same content. I mean, people have always ****ed with their settings anyway, industry standards be damned (D65 is 'too warm', crank up the sharpness and contrast etc), so we were always looking at something different even in SDR, but HDR introduced a whole 'nuther level of cluster****iness to that equation, royally screwing with the user implementation before anyone even touched a dial to adjust it this way or that.

PS

I love HDR to bits, the above is not a hit piece on what it does as a system in itself, just the disastrous consumer implementation of it.

[Fjodor2000]

Originally Posted by nissling

Again this is a completely irrelevant question for the reasons I already stated. The intent is what Östlund had in mind, and that can practically speaking be "whatever". It may be brighter, it may be the same, it may be darker. I don't know.

With this kind of reasoning, the video in all movies is already perfectly represented on the released (4K) BD as intended within the boundaries of the used format, if director approved. No need for any analysis or discussion about it. End of story.

[Fjodor2000]

Originally Posted by dorian

That isn't true. The maximum output is dictated solely by the display.

We're of course talking about when viewing the movie on a display capable of representing the video on the disc.

Otherwise everything is up in the air. What if you watch the movie on a TV only capable of displaying 50 nits, and only 50% of the Rec.709 color space? Should you use that as reference, or even take it into account, when judging the video quality on the (4K) BD?

Originally Posted by dorian

Nissling went over this but it's worth reiterating that there isn't really such a thing as HDR scanning.

No need to derail the thread further. I've already clarified what I meant by the difference between outputting a master in SDR vs HDR format (and the limitations in e.g. max brightness depending on format chosen), and that scanning is only a part of the process.

Originally Posted by dorian

Why is representing reality in an absolute fashion the artistic goal?

Never said it was. Did you not read this? :

Originally Posted by Fjodor2000

You can of course argue that there is always the creator's intent. For example perhaps the director doesn't want it to look bright, not even the frame average brightness in any of the scenes, regardless if shot under very bright conditions. That's totally fine.

[nissling]

Originally Posted by Fjodor2000

With this kind of reasoning, the video in all movies is already perfectly represented on the released (4K) BD as intended within the boundaries of the used format, if director approved. No need for any analysis or discussion about it. End of story.

You are taking me way too literally.

I like to analyze UHDs (and regular BDs as well) because it gives me a better understanding of how a certain film has been treated. They vary greatly and in many cases I'm surprised to see just how the objective findings completely contradicts other people's subjective claims. It's interesting and something I like to spend some of my free time on. I don't review the releases though, as in giving them scores, but I can sometimes give a certain release a recommendation if I find it to be something extraordinary.

There are of course objective parameters and facts. Too much noise reduction cases loss of detail, too much sharpening causes halos, too much clipping causes magnificent loss in highlight coverage et cetera, which does in practice reduce image quality. Any director is of course free to use them for their work but that doesn't make them look good (Cameron...).

In the case of Triangle of Sadness, if you have any opinions regarding the approved grade feel free to express them. I think it looks spectacular and if it's the way the director wants it then the Criterion is a perfect choice in my world, but others are of course free to dislike it.

[Macatouille]

Brick (2005)

Mastering display luminance              : min: 0.0001 cd/m2, max: 1000 cd/m2
Maximum Content Light Level              : 336 cd/m2
Maximum Frame-Average Light Level        : 297 cd/m2

Heatmaps (Album)

[Show spoiler]

Gamut Visualizations (Album)

[Show spoiler]

Tonemapped Screenshots (Album)

[Show spoiler]

Plots (HDR10, Dolby Vision, Bitrate)


Past Posts (Fully updated, as of this post)

[Macatouille]

Fairly conservative luminance values here: peak highlight at 420 nits, average frame at 313. Pretty much in line with past Johnson/Yedlin endeavors. When I compare this disc to the old Kino Blu-ray I see virtually no new information in the highlights (Blu-ray, raw 4K, tonemapped 4K):









All that being said, the WCG usage is surprisingly strong here, extending into Rec.2020 in the blues:



But some reds show up too:

[ko8ebryant24]

Originally Posted by Macatouille

Brick (2005)

Mastering display luminance              : min: 0.0001 cd/m2, max: 1000 cd/m2
Maximum Content Light Level              : 336 cd/m2
Maximum Frame-Average Light Level        : 297 cd/m2

Heatmaps

[Show spoiler](Album)

[Show spoiler]

Gamut Visualizations (Album)

[Show spoiler]

Tonemapped Screenshots (Album)

[Show spoiler]

Plots (HDR10, Dolby Vision, Bitrate)


Past Posts (Fully updated, as of this post)

Nice work! Are these from the base layer?

[Macatouille]

Originally Posted by ko8ebryant24

Nice work! Are these from the base layer?

You mean the heatmaps and gamut visualizations? Yep. I rip the disc with MakeMKV and use ffmpeg for the screenshots, specifically this command:

ffmpeg -i "C:\Drive\Movie.mkv" -vf fps=1/30 C:\Drive\Movie\Movie%04d.png

That takes a raw screenshot every 30 seconds, names them sequentially, and puts them in a directory. I then go through each, deleting any duplicates and those that aren't sharp. After that I put each of them through Dorian's two scripts, which I can link to if you can't find them. I have adapted them for offline usage so I don't have to upload the images to Google.

I also run this command to get the tonemapped version of each screenshot above:

ffmpeg -i "C:\Drive\Movie.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:\Drive\MovieTonemapped\MovieT%04d.png

I use MediaInfo to grab the metadata and DoVi_Scripts for the plots. ImgBB seems to be a good host for the images, at least so far (they haven't deleted anything).

[TbeRw01]

Originally Posted by Macatouille

After that I put each of them through Dorian's two scripts, which I can link to if you can't find them.

FYI, Dorian's scripts are implemented in DoV_Scripts Workflow 7-3. This workflow automatically export HDR png with FEL baking support and create the heatmap and gamut visualization. The script can also batch process a folder with PNG.

By default, it will export 25 frames with a 2000 frames interval but you can change that at lines 135-137

:: choose the amount of frames to export in 7 (default=25)
set frame_number=25
:: choose the frame interval for the screenshot export in 7 (default = 000 which means every 2000 and 00 would mean every 200 etc...)
set frame_interval=000

[Macatouille]

Originally Posted by TbeRw01

FYI, Dorian's scripts are implemented in DoV_Scripts Workflow 7-3. This workflow automatically export HDR png with FEL baking support and create the heatmap and gamut visualization. The script can also batch process a folder with PNG.

By default, it will export 25 frames with a 2000 frames interval but you can change that at lines 135-137

That's good to know.

Does that mean it'll export 25 frames in a row, pause for 2000 frames, export 25 more, pause, etc. Or do you mean it'll export 1 frame, pause for 2000 frames, export another, and then repeat until it hits 25 total?

[TbeRw01]

The script will first export all the image, then create all the heatmap and then the gamut and then will switch to a manual mode so you can export a specific frame.

so while it exports the png, you can go in the folder and delete the out of focus ones to save time.

[mrtickleuk]

Originally Posted by TbeRw01

FYI, Dorian's scripts are implemented in DoV_Scripts Workflow 7-3. This workflow automatically export HDR png with FEL baking support and create the heatmap and gamut visualization. The script can also batch process a folder with PNG.

Nice, sounds like cookery. ON that note - happy Chocolate Cupcake Day everyone!
https://www.daysoftheyear.com/days/c...e-cupcake-day/