HDR heatmaps and torch mode UHDs

[Geoff D]

Originally Posted by andreasy969

IMO there's actually nothing wrong with going beyond 1.000 in general, if used properly as in for very small highlights only. Since perceived brightness (brightness is the perception) doesn't increase linear with increased luminance (the factual value), 10.000 nits also doesn't (or shouldn't - never seen 10.000 nits ) look 10 times brighter than 1.000 nits. (which I think is also the reason why the lower nits values come with much higher precision btw)

1.000 nits APL is another story though - "The Meg" says hello.

Yep yep yep, the notion that HDR is constantly pumping out at thousands of nits and can damage eyes is one of the hoariest possible myths about the format. I've been enjoying a TV that can hit ~1800 nits peak for the last three years, while not all content will get near that there's been no deterioration in my eyesight whatsoever, as evinced by the eye test that I had just three weeks ago.

[Kris Deering]

I've been thinking a lot about nits and highlights and such these last few weeks while working on a project. Information out there suggests that the human eye at normal to high bias has a perceptible contrast ratio of about 100:1. With a relatively average to high APL scene, you really don't need a strong highlight to hit that range, so highlights in general don't really need to be that high to be effective. Our contrast perception goes up dramatically as the average APL goes a lot lower (think Game of Thrones episode that people *****ed about). The increase in contrast perception helps us tremendously with seeing details near black (which also takes A LOT more sequential contrast potential from the display as well), so one again high nit value highlights are not really needed that much because the optical comparison of the two values would help.

The biggest benefit for high luminance in HDR is color volume (luminance+saturation). There are just some colors that are not achievable without the luminance capability of HDR (and a display that can do them). We already see the byproducts of not being able to do them with desaturation issues with some content (even on the grading side when you look at Stacey's comments in his thread for his montage).

I may be inclined to agree that from a specular highlight position, 1000 nits was probably more than enough. But for overall color volume, it may not be. Though one would have to figure out what luminance level is required to meet perceptibility for color volume and how often are any colors that need that amount of luminance to be display properly ever occurring in any content.

[lgans316]

Originally Posted by andreasy969

IMO there's actually nothing wrong with going beyond 1.000 in general, if used properly as in for very small highlights only. Since perceived brightness (brightness is the perception) doesn't increase linear with increased luminance (the factual value), 10.000 nits also doesn't (or shouldn't - never seen 10.000 nits ) look 10 times brighter than 1.000 nits. (which I think is also the reason why the lower nits values come with much higher precision btw)

a little bit more verbose with my post.

1.000 nits APL is another story though - "The Meg" says hello.

I should have been verbose with my last post.

Should have stressed higher APL throughout and too much of supercharged flashing lights. Few examples I can think of that almost gave me an eye ache on my Panny DX902.

1. MMFR storm scene
2. BvS - fight between B and S and the trio fighting against Doomsday.
3. Godzilla KOTM - Nuclear Godzilla Vs. Monster Zero last scene.

[andreasy969]

Originally Posted by lgans316

I should have been verbose with my last post.

Should have stressed higher APL throughout...

All good. I actually think maybe I should've been less pedantic. Wasn't completely sure what you meant though.

[Fjodor2000]

Originally Posted by Pyoko

As for the numbers, in that particular case if you were to play the BD, and you set your display to gamma 2.4 and maximum white to output 285 nits, then you would get something that roughly resembled the UHD in average brightness on the low-to-mid end, except the UHD would also have much brighter highlights on top of that because that's what HDR is about.

So if you could attenuate the highlights of the UHD to the same degree as they are on the BD (this is what is meant by highlight roll-off), which is something completely theoretical and not something you can or would want to do, then they would look similar with those settings.

I'm trying to understand this, so please bear with me if I get it wrong. Anyway, from this example:

With a similar highlight roll-off the HDR version would on average be equivalent to the SDR version with peak white at 238 nits (Gamma at 2.40), based on 11 samples. (Min: 139, Max: 320, Median: 246)

My interpretation is that the HDR has on average been master brighter than the SDR on the BD. Is this correct?

Also, does it mean that the HDR version should have lowered the average brightness so the 238 nits value mentioned above would be 100 nits instead, to reach the same average brightness level as the SDR on the BD? This would be due to that SDR is aiming for 100 nits for peak brightness.

Finally, is it always desired to aim for that when mastering HDR brightness? Is there any reference level defined for HDR in this regard?

[Pyoko]

Originally Posted by Fjodor2000

My interpretation is that the HDR has on average been master brighter than the SDR on the BD. Is this correct?

Yes, compared to the BD with peak white at 100 nits. That value is always a relative measure. Occasionally you might get a BD that was mastered unduly dark, in which case it might seem like UHD was very bright when in fact it was at sensible levels and it was the BD that was anomalous. As such the numbers make most sense when comparing UHDs/BDs from the same master.

Also, does it mean that the HDR version should have lowered the average brightness so the 238 nits value mentioned above would be 100 nits instead, to reach the same average brightness level as the SDR on the BD? This would be due to that SDR is aiming for 100 nits for peak brightness.

Yes, if the analyzed value is around 100 nits that means the UHD was mastered similar to the BD at the 100 nits reference level.

Finally, is it always desired to aim for that when mastering HDR brightness? Is there any reference level defined for HDR in this regard?

I believe there are two reference levels for HDR, the older one is SDR-based having "diffuse white" at 100 nits, the other has diffuse white at 200 nits, which might explain why so many, but not all, UHDs are brighter. But between the camera exposure, lighting, and artistic grading decisions I don't know if they're actually keeping track of where this diffuse white would be on the content while they're grading it. Right now it seems like it's mostly a free-for-all.

Obviously you could master brighter or darker on SDR as well, however the available range there is so small that unless you really go crazy, you're likely to end up with something sensible and fairly similar in levels compared to all the other SDR stuff out there. The much greater range of HDR means there's also a much greater range in the average brightness across titles.

[Kris Deering]

Originally Posted by Pyoko

Yes, compared to the BD with peak white at 100 nits. That value is always a relative measure. Occasionally you might get a BD that was mastered unduly dark, in which case it might seem like UHD was very bright when in fact it was at sensible levels and it was the BD that was anomalous. As such the numbers make most sense when comparing UHDs/BDs from the same master.


Yes, if the analyzed value is around 100 nits that means the UHD was mastered similar to the BD at the 100 nits reference level.


I believe there are two reference levels for HDR, the older one is SDR-based having "diffuse white" at 100 nits, the other has diffuse white at 200 nits, which might explain why so many, but not all, UHDs are brighter. But between the camera exposure, lighting, and artistic grading decisions I don't know if they're actually keeping track of where this diffuse white would be on the content while they're grading it. Right now it seems like it's mostly a free-for-all.

Obviously you could master brighter or darker on SDR as well, however the available range there is so small that unless you really go crazy, you're likely to end up with something sensible and fairly similar in levels compared to all the other SDR stuff out there. The much greater range of HDR means there's also a much greater range in the average brightness across titles.

I'm not aware of any 4K HDR content that has been mastered with reference white at 200 nits. This was brought up as a way to compensate for the fact that most people feel like HDR is too dark on a flat panel (because they were watching 100 nit grades WAY higher than intended).

Also, ADL for a HDR title should be LOWER than a SDR grade. You're spreading the grade out over much more of a range, whereas SDR is more compressed in brightness. This has been shown many times with analysis done on 4K HDR frames of a movie when compared directly to the SDR grade using grading tools for analysis. Peak levels are definitely brighter, but ADL tends to be lower.

[Pyoko]

Originally Posted by Kris Deering

Also, ADL for a HDR title should be LOWER than a SDR grade. You're spreading the grade out over much more of a range

I don't follow why this would be. The way I see it, unless the SDR had a brightened low-range to avoid crushing/bitdepth issues, the HDR average should be brighter. Trying to reproduce the same full-range linear scene at the same "exposure" both would be the same up until the clipping point of the SDR, at which point the HDR would just add even more light on top of that. And if the highlights on the SDR were compressed/rolled off that would bring the average down even further because the compression curve has to begin darkening things down a fair bit down in the values to be smooth.

This has been shown many times with analysis done on 4K HDR frames of a movie when compared directly to the SDR grade using grading tools for analysis. Peak levels are definitely brighter, but ADL tends to be lower.

Well it certainly hasn't been shown here, out of two-dozen discs I've checked only Alita and The Matrix are darker in HDR, and that's ignoring the HDR highlights completely.

[Fjodor2000]

So how do we determine if an UHD movie has been well graded and mastered for HDR from one that hasn't?

What number should we look for in the analyses provided in this thread?

[Pyoko]

Well that's subjective and depends on too many factors to give a simple answer that always applies, but in my opinion between 100-150 nits relative to the Blu-ray is good (provided the BD wasn't unduly dark.) After that it starts getting needlessly bright though probably still fine and perhaps not even very noticeable in many cases, especially if it's a darker movie overall. Stuff like Starship Troopers or Close Encounters is just really unnecessary and counterproductive though.

As for native HDR vs SDR upconversions it's more difficult to explain, but you can often tell to some extent when you look at the heatmaps. Here are a couple of examples from Black Hawk Down. To the left is the actual HDR from the disc, to the right I've simulated an SDR upconversion by first converting it to SDR, and then trying to expand that back into HDR using an arbitrary curve.





Even though most of the highlight detail is probably still in there somewhere, you can tell the variation in the highlights has been lost, as has the scene linear rise in brightness. Things that shouldn't be all that bright have gotten caught up in the HDR expansion curve and the upper end consists of more flatter areas with samey values rather than terminating in finer details.

[Streetlight]

Any comparison for color space between Mastered in 4K BD and UHD BD?

[ARROW-AV]

Originally Posted by Pyoko

Pyoko thank for you this. Absolutely superb!

Does there exist any particular software that can be used to analyse either a movie frame, or even better a whole movie, and thereby produce Gamut Triangles showing the corresponding color gamuts coverage like this?

.

[Pyoko]

I think this is the first catalog DI title I've seen that may have had an HDR grade redone from source. It falls somewhere between obvious SDR upconversion and true uncapped HDR. It has highlight detail that plain isn't there on the SDR Blu-ray and a luminance distribution that's impossible to replicate with just a curve on top of the SDR, but still with a marked highlight compression down to 200-500 nits for the most part. Either it was redone but kept conservative, or it was converted but they had a master that was less clipped than the BD one and they did a pretty good job of expanding the highlights.

Regardless of the case it's a good upgrade in terms of unclamping the highlights, with a nice uptick in sharpness/detail/grain as well. The one thing I didn't love was the grade which frequently had a lot of color sucked out of it compared to the Blu-ray, with skin tones in particular looking unnatural, as if selective desaturated. It works with the tone of the film especially as it progresses, but looks a bit weird in the beginning of the movie:










Album

With a similar highlight roll-off the HDR version would on average be equivalent to the SDR version with peak white at 203 nits (Gamma at 2.40), based on 23 samples. (Min: 119, Max: 314, Median: 198)

[Pyoko]

Wasn't quite sure what to expect from this but it's for sure massively superior to the BD. Not only is the resolution and sharpness much higher, but the BD compression is godawful often leading to detail loss worse than the lower definition. Even the UHD has trouble keeping up with all the little details and particles and fine textures occasionally.

The HDR doesn't always have a ton of information in the brightest highlights owing to the animation/rendering style, but it's a lot more than exists on the SDR grade and looks great when it's there. The BD also has some detail loss due to saturation clipping/compression, which you can see in the second shot.

UHD color | UHD luminance | BD luminance | UHD tonemap | BD upscale

With a similar highlight roll-off the HDR version would on average be equivalent to the SDR version with peak white at 187 nits (Gamma at 2.40), based on 20 samples. (Min: 139, Max: 234, Median: 197)

[Pyoko]

Predator in Predatorvision. It's a shame P3 and Rec709 share the blue primary because it ruined the gamut-rendering of the heatvision shot.

If you quickly flip between the HDR and SDR of the explosion, you almost get a stereoscopic effect.









Album

With a similar highlight roll-off the HDR version would on average be equivalent to the SDR version with peak white at 144 nits (Gamma at 2.40), based on 20 samples. (Min: 47, Max: 469, Median: 118)

[Eric_The_Victor]

Yeah Predator shot number 6 looks awesome! Too bad more frames in more movies don't go out to P3 or rec2020. I'll take wider color over more nits. I see you own Evil Dead 2, can you show us that movie please? I feel it's a little overcooked vs the first movie but what to see how the gamut is.

[BrownianMotion]

The Great Wall has some of the best looking colors I've seen. I wonder what the gamut analysis on that would look like.

[Amano]

The predatorvision is a good showcase of the additional P3 pallette If graded with rec2020 capabilities in mind it could have been completely free of rec709 colors.

In sum: I like what they did to this movie. That feels "final".

[Pyoko]

Originally Posted by Eric_The_Victor

I see you own Evil Dead 2, can you show us that movie please? I feel it's a little overcooked vs the first movie but what to see how the gamut is.

Will try to do Evil Dead 2, wanted to check it myself as well.

Originally Posted by Amano

The predatorvision is a good showcase of the additional P3 pallette If graded with rec2020 capabilities in mind it could have been completely free of rec709 colors.

Probably yeah, it's definitely an interesting shape on the gamut.

[Pyoko]

The movie itself doesn't have a huge amount of shots that could really show off any crazy HDR, and no color to speak of, but the UHD is fantastic compared to the horrible, clipped BD.

I did notice some kind of grain/posterization error that popped up here and there. Most of the time it was really subtle, this was the worst shot (mainly the back of the prostitute, but the grain on the white wall also looks a tad funky):


Also that last shot looks fabulous in black and white.









Album

With a similar highlight roll-off the HDR version would on average be equivalent to the SDR version with peak white at 150 nits (Gamma at 2.40), based on 26 samples. (Min: 75, Max: 383, Median: 137)