AVS Forum banner

11461 - 11480 of 11613 Posts

·
Registered
Joined
·
8,142 Posts
Since it looks like 'Display Peak Luminance' and 'Dynamic Target Nits' will stay with us at least for for quite a while, would it be much work to include them in the assignable shortcuts section? I believe it would greatly help balance them out in real time with running content.
I only have very little time atm, so I'd rather spend the available time on improving the algos atm.

Maybe there is a trick to make it work? Only desaturating much more aggressively near the very brightest pixels?
That's what desat 2-4 and dumb desat do.

These artifacts are only of one "solid" color it seems?
The inside of an explosion might also have a "solid" center. I'm not sure we can use that as a deciding factor. It's also not too easy to detect.

Moreover, with "max" the artifact is even darker than in the non-tone-mapped image(?). How is that possible?
That's an optical illusion... :D Try a pixel peeper. E.g. I'm using this freeware:


Well it would be more like mixing desat4 with desat1 actually, but desat4 is already mixed with desat1, right? The problem is that we can not control how much desat4 and how much desat1 are "mixed".

The idea would be to keep more desat1 than desat4 (using desat1 to desaturate pure colors a bit more). Does that make sense?
Do you really mean desat1? Desat1 is part of the brightness adjustment, as recommended by BT.2390. Basically the more I compress the luminance of a pixel, the more desaturation I apply. That's desat1.

Desat2-4 and dumb desat all work completely different. I'm not sure if it makes sense to make desat1 stronger. I don't really think so, to be honest. I tried that once but it looked really bad. I think it's a much better approach to add desat2-4 and/or dumb desat on top of desat1.

If you use a mixture of 0, you get desat2. A mixture of 50 is desat3. A mixture of 100 is desat4 or dumb desat. Did you really mean desat1? Or did you mean desat2? I'm not completely sure. In either case, I'm still not fully understanding the reasoning behind your idea. Why do you want to use more of desat1 (or maybe desat2)? Why do you think that would help? I'm not saying I won't offer that, I'm just wondering what the logical reason is why you're asking for that? Can't follow your chain of thought here atm.

there have been a few titles(namely starship troopers and king kong2005) which have had an uncomfortably bright appearance, and I found I needed to adjust dpl(say from 50 to 150 or 200) to provide a comfortable viewing level, just increasing dtn previously didnt resolve this. i think dont add peak + the higher dtn + max lum method resolved it, and possibly the recovery assisted.
Ah ok. So you're saying the latest build works better for you for some reason?

Yo, can we settle on a (Scientifically Correct DTM Tab) and a (Experimental DTM Tab) to split the settings & testing?
Sorry, but no. GUI design costs a LOT of time. And doing this for options which are only supposed to stay in the short term does not seem like a wise investment to me.

Can custom curves fix/improve this?
I've no idea.

Generally, a key question here is: Do we absolutely have to have perfect detail, on the cost of saturation? Or do you prefer to have a bit more saturation in highlights and buy that by losing some detail? Looking at test patterns, it's tempting to aim for more detail. But in real movie scenes that's not always what looks best. So I don't think there's an easy answer here.

I would prefer using test patterns only to double check things and to look for clues of problematic areas etc. But I would prefer making decisions more based on real movie content instead of test patterns. I do love test patterns, though, and like to use them extensively.
 

·
Registered
Joined
·
6,451 Posts
@madshi
Since it looks like 'Display Peak Luminance' and 'Dynamic Target Nits' will stay with us at least for for quite a while, would it be much work to include them in the assignable shortcuts section? I believe it would greatly help balance them out in real time with running content.
You can accomplish this to a great extent just using multiple profiles and switching between them via shortcut.
 

·
Registered
Joined
·
2,677 Posts
Ah ok. So you're saying the latest build works better for you for some reason?
Yes, more consistent between titles.

Also, I noticed I like the way stronger desat looks more with contrast recovery, but I wasn't a fan of contrast recovery as it flattened out the image(when in motion at least), the only adjustment that had any effect was dpl, by reducing it to half of what I was using the depth came back.

That dpl would be what you measure off the screen with the meter closer to the seating position. i.e. say you're watching 100nits sdr, you measure 50 nits off the projector with the meter at the screen pointed at the projector using the diffuser, but if you measured with the meter pointed at the screen and closer to the seating position you'd only measure ~25nits. Using that figure I can use contrast recovery.

This has provided the most balanced natural image so far, there were always instances where faces or skin in general stood out in the frame, either looking too dark in stronger lighting, or too bright for the lighting, giving a glow to skin or a cgi/green screen look.

Sorry anecdotal, and something that only helps projection, and something I've had trouble screenshotting or photographing. wanted to mention in case it helps anyone. I can use a stronger desat without issue, fire looks like fire, and the image looks natural and balanced. went through a myriad of content and it worked with everything, from total recall to the hobbit, tremors to solo

31.png
 

·
Registered
Joined
·
158 Posts
We could use other SDR Blu-ray titles for skin tones, but how do we know if one was an automated trim pass or one that was manually supervised by the colourist? Or how close were the two grades? (...) And then they proceed to give a list of options for colourists to make manual adjustments to the image produced by the SDR trim pass. How many of them do this though, and to what degree?
Those are excellent questions for which there's no simple answer. I'm sure there are some colorist/DP/Directors out there pedantic enough to go through every version of their movie with fine-tooth comb to ensure they're all as close to perfect as possible. But my gut feeling, based on what I know of human (and corporate) nature, tells me that the cheapest and least time-consuming option is, most of the time, a preferred one. IMO unless we have a statement from someone involved in the production, it's probably safer to assume that the path of the least resistance was taken.
 

·
Registered
Joined
·
822 Posts
@Javs can you please reach out to anyone you know who did the HDR and SDR grades on a film, who you trust to have done a good job, then we use that for testing? The other option is to wait for the new Spears disc, but who knows when that will be ready.

It seems like we will never get people to settle on this until we have a reference point we can trust.
 

·
Registered
Joined
·
6,451 Posts
^^^ @bdht I think it's worth taking a step back and instead of looking at it just as a 25/75 DPL/DTN combo, with "don't add peak nits" checked, let's see what the resulting FALL multiplier is. Because I think what it seems like 25 nits DPL is doing for you is maybe really that you just like that resulting multiplier.

Here are the three formulas I use for these purposes (always assuming "don't add peak nits" is checked):

DTN = multiplier * 50 / log10(DPL)
multiplier = log10(DPL) * DTN / 50
DPL = 10^(multiplier * 50 / DTN)

So using the second formula, your resulting multiplier is 2.097. You can then use that same multiplier and your actual real peak nits in the first formula and get a DTN to use with that real DPL that will result in the same multiplier and thus same targets.

So say your real DPL is 50 nits, it gives you a DTN of 61.7. We can't use decimal DTNs, but 50/62 combo gives a multiplier of 2.107, which is very close, and if you wanted to be really anal and get it even closer, you could run 49/62 combo, which gives a multiplier of 2.096.

These combos should give you the same targets you are loving, just with the higher DPL.
 

·
Registered
Joined
·
2,677 Posts
^^^ @bdht I think it's worth taking a step back and instead of looking at it just as a 25/75 DPL/DTN combo, with "don't add peak nits" checked, let's see what the resulting FALL multiplier is. Because I think what it seems like 25 nits DPL is doing for you is really that you just like that resulting multiplier.

Here are the three formulas I use for these purposes (always assuming "don't add peak nits" is checked):

DTN = multiplier * 50 / log10(DPL)
multiplier = log10(DPL) * DTN / 50
DPL = 10^(multiplier * 50 / DTN)

So using the second formula, your resulting multiplier is 2.097. You can then use that same multiplier and your actual real peak nits in the first formula and get a DTN to use with that real DPL that will result in the same multiplier and thus same targets.

So say your real DPL is 50 nits, it gives you a DTN of 61.7. We can't use decimal DTNs, but 50/62 combo gives a multiplier of 2.107, which is very close, and if you wanted to be really anal and get it even closer, you could run 49/62 combo, which gives a multiplier of 2.096.

These combos should give you the same targets you are loving, just with the higher DPL.
thanks ill give it another try, i had tried adjusting dtn down with 50 nits dpl and contrast recovery enabled, but it often had no effect on the frame, whereas dpl would.

i think maybe that lower starting point is whats balancing contrast recovery.
 

·
Registered
Joined
·
6,451 Posts
thanks ill give it another try, i had tried adjusting dtn down with 50 nits dpl and contrast recovery enabled, but it often had no effect on the frame, whereas dpl would.

i think maybe that lower starting point is whats balancing contrast recovery.
The biggest difference between the two should be on scenes with peaks or targets between 25-49 nits, where the 49/62 combo will still map 1:1 while the 25/75 combo will start compressing.

I'd create two profiles that are identical other than the 25/75 vs 49/62 DPL/DTN, and switch between them.
 

·
Registered
Joined
·
89 Posts
You can accomplish this to a great extent just using multiple profiles and switching between them via shortcut.
Thanks for the hint, I actually thought of that as well, but that would take a ton of profiles, since it is about finding the sweetspot between these two settings.

Also, as explained, I would say these two setting belong to the essential set meanwhile, and all the essential settings have assignable keys, so it would make sense including these two.

Gesendet von meinem ELE-L29 mit Tapatalk
 

·
Registered
Joined
·
6,451 Posts
Thanks for the hint, I actually thought of that as well, but that would take a ton of profiles, since it is about finding the sweetspot between these two settings.
Yes, I have 5-10 profiles set up regularly. Does that qualify as a ton? :) It works really well for pretty much any kind of comparison testing I'm trying to do.
 

·
Registered
Joined
·
146 Posts
^^^ @bdht I think it's worth taking a step back and instead of looking at it just as a 25/75 DPL/DTN combo, with "don't add peak nits" checked, let's see what the resulting FALL multiplier is. Because I think what it seems like 25 nits DPL is doing for you is maybe really that you just like that resulting multiplier.

Here are the three formulas I use for these purposes (always assuming "don't add peak nits" is checked):

DTN = multiplier * 50 / log10(DPL)
multiplier = log10(DPL) * DTN / 50
DPL = 10^(multiplier * 50 / DTN)

So using the second formula, your resulting multiplier is 2.097. You can then use that same multiplier and your actual real peak nits in the first formula and get a DTN to use with that real DPL that will result in the same multiplier and thus same targets.

So say your real DPL is 50 nits, it gives you a DTN of 61.7. We can't use decimal DTNs, but 50/62 combo gives a multiplier of 2.107, which is very close, and if you wanted to be really anal and get it even closer, you could run 49/62 combo, which gives a multiplier of 2.096.

These combos should give you the same targets you are loving, just with the higher DPL.
I understand the FALL multiplier is subjective, but is there a range you would recommend?

My DPL is 124 nits. I've already experimented with all sorts of DTN values, but would be nice to have a more scientific approach to determining the correct multiplier, given we have these precise formulas.

I do think the "don't add peak nits" has been the biggest improvement I've seen out of these recent beta releases.
 

·
Registered
Joined
·
1,339 Posts
That's what desat 2-4 and dumb desat do.
Well your max2-4 approach is far more effective to remove the blue artifacts ;)
For the green spear, it just need a little help. It is alright for me with "lum3" for instance, but not with "lum4".

Do you really mean desat1? Desat1 is part of the brightness adjustment, as recommended by BT.2390. Basically the more I compress the luminance of a pixel, the more desaturation I apply. That's desat1.

Desat2-4 and dumb desat all work completely different. I'm not sure if it makes sense to make desat1 stronger. I don't really think so, to be honest. I tried that once but it looked really bad. I think it's a much better approach to add desat2-4 and/or dumb desat on top of desat1.

If you use a mixture of 0, you get desat2. A mixture of 50 is desat3. A mixture of 100 is desat4 or dumb desat. Did you really mean desat1? Or did you mean desat2? I'm not completely sure. In either case, I'm still not fully understanding the reasoning behind your idea. Why do you want to use more of desat1 (or maybe desat2)? Why do you think that would help? I'm not saying I won't offer that, I'm just wondering what the logical reason is why you're asking for that? Can't follow your chain of thought here atm.
Of course we do not want a make desat1 stronger, that would be awful.
To make it more simpler: the idea was to take the result we get with "dumb" and only desaturate the pure colors from here, leaving the mixed colors alone.
But seeing how the target impacts the balance of the desaturation on some hues, I do not think it would work (better) for now. This issue is also visible with real content, mostly with the red color.

Generally, a key question here is: Do we absolutely have to have perfect detail, on the cost of saturation? Or do you prefer to have a bit more saturation in highlights and buy that by losing some detail? Looking at test patterns, it's tempting to aim for more detail. But in real movie scenes that's not always what looks best. So I don't think there's an easy answer here.
Personally, I aim for accurate skin tones first, and then for the most saturation I can get without losing details.
Maybe the custom curves can help with this. Or at worse you will have to provide several options I guess: at least one to preserve details, one to preserve saturation, and a balanced one :)
I am sure the next Spears & Munsil UHD HDR Benchmark will help us a lot to fine-tune all this.

------------------------------------------------------------------------------------------------------------------------------

What is happening here with "max" around all the clouds? :eek:

non-tone-mapped / @ 100nits + sep 1.000 50-175 / @ 100nits + max 1.000 50-175

This is really awful. Back to sep for me :oops:
 

·
Registered
Joined
·
9,113 Posts
Personally, I aim for accurate skin tones first, and then for the most saturation I can get without losing details.
Maybe the custom curves can help with this. Or at worse you will have to provide several options I guess: at least one to preserve details, one to preserve saturation, and a balanced one :)
I am sure the next Spears & Munsil UHD HDR Benchmark will help us a lot to fine-tune all this.
I agree completely.
------------------------------------------------------------------------------------------------------------------------------

What is happening here with "max" around all the clouds? :eek:

non-tone-mapped / @ 100nits + sep 1.000 50-175 / @ 100nits + max 1.000 50-175

This is really awful. Back to sep for me :oops:
Does it do this with BT2020? Really need to make that distinction, if there is a global Rec709 issue - which I think there is since I saw it clearly on Atomic Blonde, then I would ask that you just note it in your mind and do as much testing in BT2020 as you can when you see issues like that since now you are looking at two variables, and perhaps Madshi will need to look specifcally at a fix for the gamut conversion after we get past this. ?
 

·
Registered
Joined
·
1,339 Posts
Does it do this with BT2020? Really need to make that distinction, if there is a global Rec709 issue - which I think there is since I saw it clearly on Atomic Blonde, then I would ask that you just note it in your mind and do as much testing in BT2020 as you can when you see issues like that since now you are looking at two variables, and perhaps Madshi will need to look specifcally at a fix for the gamut conversion after we get past this. ?
Yes it does. I am not talking about the Rec709 issue here with the orange cloud at the center, which is another thing.
Look at the clouds at the top the image and at their "edges". There is like a blue/violet line all around (and I am not using any highlight recovery here).
I do not know how I missed this before o_O
 

·
Registered
Joined
·
9,113 Posts
Yes it does. I am not talking about the Rec709 issue here with the orange cloud at the center, which is another thing.
Look at the clouds at the top the image and at their "edges". There is like a blue/violet line all around (and I am not using any highlight recovery here).
I do not know how I missed this before o_O
I see it.

Do you have Highlight Recovery off in these shots?

I am actually amazed that you use highlight recovery, I could show you artefacts all day long from that.
 

·
Registered
Joined
·
1,339 Posts
I see it.

Do you have Highlight Recovery off in these shots?

I am actually amazed that you use highlight recovery, I could show you artefacts all day long from that.
You quoted me saying that I was not using any highlight recovery :p
But I agree with you, it produces artifacts and it would be better to find another way (custom curve?) to get those highlight details back.
 

·
Registered
Joined
·
9,113 Posts
You quoted me saying that I was not using any highlight recovery
But I agree with you, it produces artifacts and it would be better to find another way (custom curve?) to get those highlight details back.
Sorry yes you are right. I'm dead tired.

Only really able to pop in here and read.
 

·
Registered
Joined
·
490 Posts
I think there are issues with grain with max.

We've been busy looking at digital sources and forgot about film. Sigh.

There are probably loads of scenes like this....


At 500 nits (La La Land) frame 69263



sep vs max




edit:

max and sep mix has it to a degree




At 1300 nits (Black Hawk Down)



sep vs max




At 400 nits (La La Land)



sep vs max




At 980 nits (Black Hawk Down)


sep vs max



@Neo-XP , do you have that Great Gatsby scene sample that you used in the past? Same thing with sep vs max there?
 

·
Registered
Joined
·
6,451 Posts
I understand the FALL multiplier is subjective, but is there a range you would recommend?

My DPL is 124 nits. I've already experimented with all sorts of DTN values, but would be nice to have a more scientific approach to determining the correct multiplier, given we have these precise formulas.
Sorry, but I think as you've probably noticed just trying to hone in on your preferred DTN value, there is no "correct" DTN value and resulting multiplier. It varies so much from person to person, and even for the same person from frame to frame. As great as the current targeting algo works (and the avgHL ceiling and sky detection are both potentially beneficial parts of this), I feel that it is an area where a lot more improvement could still be made, and that's why I have been experimenting with the avgHL ceiling and asked for it to be tunable. But the current focus in this thread is on lum, desat, etc., and hopefully we can get that squared away soon enough.

I do think the "don't add peak nits" has been the biggest improvement I've seen out of these recent beta releases.
So what's kind of interesting is I absolutely love "don't add" on high DPLs, but when I did some (admittedly very limited) testing on low DPLs I thought maybe I preferred it adding. But it wasn't anywhere near enough testing to draw any real conclusions, so we can just dismiss it as nonsense. :)
 
11461 - 11480 of 11613 Posts
Top