AVS Forum banner

6621 - 6640 of 9410 Posts

·
Registered
Joined
·
689 Posts
Fantastic improvement. Any chance of fixing the measurement file bug or is that a longer fix that will take away from this development cycle?
 

·
Registered
Joined
·
5,350 Posts
Yep, it looks bad with HDR output here as well. :)

Good. Well....not "good" but at least it's not something in my setup that's messing it up. That means madshi should be able to find and squash it. Cause once that's done, we are in hog heaven with the improvements in these new builds! Absolutely stunning.

@madshi rolled back to 96 and the issue is not there. So this was introduced in the build 97 changes. Hopefully that helps narrow down the issue. Thanks!
 

·
Premium Member
Joined
·
727 Posts
I think the saturation with clipping point 10 has the same reason: With a clipping point of 10, some pixels are much brighter, so they aren't desaturated as much.
I am currently at 15 clipping point. I find that at the default value of 40, the perceived reduction in saturation creates a more flat image and less pop.

Btw, found an issue with the scene right after the 1st screenshot Neo-XP posted in post 6613. He says, "You meet candle egg... nice skull" haha, then they pan to Zhang and his face immediately jumps in brightness.

Goes from this:


To this:



My settings (I also removed the histogram smoothing):
 

·
Registered
Joined
·
1,780 Posts
I'm mostly liking what the HSTM is doing as far as perceived contrast enhancement, but also I am seeing what other people are calling the desaturation, particularly noticeable in skintones.

I know madshi talked about this, but I still can't help but feel I wish the skintones looked as full as they did with HSTM off.
 

·
Registered
Joined
·
1,780 Posts
Build 97c goes a bit crazy with one scene in Casino. 1:00:42 to 1:00:47 the histogram is adjusted several times.


You mean screen brightness flickering?

I thought we are supposed to ignore that now.

I’ve seen that too pretty bad in some places.

I also saw a scene where the black bars on the scope movie flashed white for a moment. That seems like a strange new artifact for me as well.
 

·
Registered
Joined
·
1,208 Posts
Btw, found an issue with the scene right after the 1st screenshot Neo-XP posted in post 6613. He says, "You meet candle egg... nice skull" haha, then they pan to Zhang and his face immediately jumps in brightness.
That is because of the initial and final power you used.

HSTM Off


The Meg - Frame 8810


HSTM On (initial power - clipping point - final power) - 10-40-10 / 20-40-10 / 30-40-10 / 40-40-40 / 50-40-50 / 60-40-60 / 70-40-70



10-40-10 is too much here IMO.

Build 97c goes a bit crazy with one scene in Casino. 1:00:42 to 1:00:47 the histogram is adjusted several times.
That is normal and explained in madshi's note to test:

Important note: The histogram shaped TM is currently not smoothed in the time domain. Which means if you use this feature for playback, there could potentially be some flickering or similar artifacts. Please disregard this for now. Of course I can smooth this later, for now the key question is whether this is an algo we want to use at all, or not. AND if this algo (if we use it) can replace sky detection, or if both should be used at the same time.
 

·
Premium Member
Joined
·
16,949 Posts
I am coming from build 87 to the newest, 97c.

Like others I am also observing flicker, as noted it is to be ignored knowing it is temporary.

Something has changed regarding color between build 87 with the 90 series. Color has noticably de saturated and the hue has changed. I had to click hue 4 points toward red to get close to build 87s default.
Highlites have noticably increased, wow what a nice improvement, crazy good. Good work madshi.
 

·
Registered
Joined
·
1,208 Posts
Something has changed regarding color between build 87 with the 90 series. Color has noticably de saturated and the hue has changed. I had to click hue 4 points toward red to get close to build 87s default.
This was explained to me by madshi here:

It all depends on the amount of luminance compression. FYI, in old builds madVR had 2 separate desaturation steps. The option "desaturate compressed pixels" disables one of those. But the other one is still active (and we need it). This still active option does desaturate pixels, depending on how bright they originally were and depending on how much they were compressed by tone mapping. Now with that in mind, please compare these 3 images:

100nits HSTM Off - | - 100nits HSTM On - | - 200nits HSTM Off

As you can see, the desaturation comes solely from the amount each pixel is compressed (= reduced in brightness by tone mapping). It doesn't come from HSTM itself. It only happens because HSTM compressed some pixels more strongly.

We do plan to re-evaluate the whole desaturation logic later. So we probably shouldn't worry about this problem at this time. For now you can rest assured that HSTM itself does not modify saturation in any way. It does modify pixel brightness, though, which then indirectly affects saturation (which is a separate algorithm outside of HSTM's control).

I think the saturation with clipping point 10 has the same reason: With a clipping point of 10, some pixels are much brighter, so they aren't desaturated as much.

Btw, honestly, the lower saturation levels look *much* more natural to me in these images. But that's a completely separate topic, of course.
Colors look more natural now, maybe they look a bit desaturated to everyone because we are used to see these scenes with the previous builds (which most probably saturated colors way too much).
 

·
Registered
Joined
·
1,208 Posts
Experiment on madshi's favorite scene:

HSTM Off


Black Hawk Down - Frame 94527


HSTM On (initial power - clipping point - final power) - 10-40-10 / 20-40-20 / 30-40-30 / 40-40-40 / 50-40-50 / 60-40-60 / 70-40-70



It looks like 20 for the initial and final power is already the sweet spot (20-40-20 has more depth than 40-40-40).

Edit: HSTM On (20-40-20) with shadow boost "medium" and "high" enabled:



Shadow boost is still relevant for shadow detail lovers.
 

·
Premium Member
Joined
·
16,949 Posts
Thank you for your post Neo-XP, thanks to everyone who contribute to the progress of Madvr, it has jumped another level in performance.
 

·
Registered
Joined
·
1,479 Posts
Discussion Starter #6,632 (Edited)
Tone mapping goal is an art.
Goal is to be as close to the source as possible:
- brightness
- color hue
- color saturation
- highlights details
- shadow details
- pop contrast

Since we can't have it all on our display with limited brightness, we have to compromise.

Now the goal is to please the MOST amount of people with the default settings of those compromise.

But they are things were people taste (it's art remember) are too strongly appart.
That's were setting come from:
- highlights recovery details
- highlights shadow details
- nice fire


For saturation vs highlights details, the choice have a huge impact on the picture.
It is never the question to have more saturation than the source. The question os how much do you want to preserve. The more you preserve, the less highlights details you get.

From what I read in the different forums around the world, most of people get a WOW effect with preserved saturation more than they do with preserved highlights details.
And I believe that despite Neo-xp and madshi who always favors highlight details, most people prefer saturation instead.

So let's keep that in mind when choosing the default desaturation settings.

Also since there are clearly strong different taste about that, it makes a lot of sense to propose a settings with maybe 3 strength: balanced, saturation lovers, highlights details lover.

It makes little sense to have a setting for a little effect like "fire hue" but not for a huge effect like " saturation vs highlights details ".
 

·
Registered
Joined
·
1,208 Posts
Experiment with the clipping point:

HSTM Off


Spears & Munsil UHD HDR Benchmark - Frame 5567

HSTM On (initial power - clipping point - final power) - 20-10-20 / 20-20-20 / 20-30-20 / 20-40-20 / 20-50-20 / 20-60-20 / 20-70-20 / 20-80-20



20-40-20 is the sweet spot here.

Therefore, the following parameters are the best for me:

 

·
Registered
Joined
·
7,948 Posts
The colors look much much more natural actually IMO:

HSTM Off / HSTM On


La La Land - Frame 40803

I mean... look at the skin color of everyone here :eek:
I agree that I like the saturation with HSTM On better, but I'm reserving final judgement for that later test build where we look at saturation once more.

Ok. Could you please provide in the next build the 96 methods keeping the same peak BUT with the change to the way the algo would adjust itself? So we can see which one really fixed the issues. :)

Also maybe you could use 96 way of doing with some limits to how bright or dark it can get: 10, 20, 30, 40... 100% change allowed from original Fall. (Probably better never to allow that the Fall drops actually).
Or respect half of the peak condition and half of the Fall but neither entirely.
Let's take it one step at a time, I don't think adding a hundred new options right now seems like the best approach. So let's first just check peak vs fall, and adjust A vs B.

What in the hell kind of wizardry is this???! I miss a few builds and skip to 97a and run a couple tests on my favorite dark scenes and OMG! What have you done? That's a HUGE improvement. I'm not looking at individual frames....I'm looking at it overall in moving video. That's insane. Like REALLY impressive! To give you an idea of what I mean, take a look at the tesseract at the beginning of Infinity War. Doing it as a paused frame is all well and good but watch it MOVING....the detail inside that thing. WOW! I'm really blown away here.
:D

Oh dear. Ummm, hey madshi...I don't think the menu on the Spears and Munsil disc is supposed to look like that. For some reason I wasn't able to take a screen cap of it. Suffice it to say I don't think it looks quite right.

I'm getting that same affect on the red background of Last Jedi in Snoke's room. I don't know what's causing it.
I'll have a look.

No smoothing at all is even better to preserve highlight details

I'll try to play with initial/final power and clipping point, but they seem extremely good already.
Oh wow, I didn't expect that. I thought smoothing would be benefitial. But we shouldn't rule smoothing out just yet. I could imagine that not using smoothing could produce artifacts in some situations. So it might still make sense to double check once in a while if smoothing 0 is *always* better than smoothing 20 or not.

It's also quite possible that setting smoothing to 0 simply increases the strength of the overall algorithm. What happens if you use a bit of smoothing but then choose a strength of e.g. 200 or 400? Does that give you a similar effect? Or is smoothing 0 still better?

Fantastic improvement. Any chance of fixing the measurement file bug or is that a longer fix that will take away from this development cycle?
It's on my list of things to look at. But fixing all the reported image quality issues comes first.

I am currently at 15 clipping point. I find that at the default value of 40, the perceived reduction in saturation creates a more flat image and less pop.
Do you have a good screenshot for this which shows the difference? As I mentioned before, saturation is a topic we will revisit later. I wouldn't want you to make decisions on HSTM settings based on saturation.

Maybe I should completely disable any desaturation for now? It will look bad, IMHO, but it should make it easier for you guys to judge which HSTM settings are better and which are worse, because saturation no longer plays a role? Or maybe, it could actually go into the other direction that you pick completely different HSTM settings because you think the saturation is otherwise too high. It's difficult... <sigh>

I'm mostly liking what the HSTM is doing as far as perceived contrast enhancement, but also I am seeing what other people are calling the desaturation, particularly noticeable in skintones.

I know madshi talked about this, but I still can't help but feel I wish the skintones looked as full as they did with HSTM off.
As mentioned multiple times, we will revisit saturation later. For now please try to ignore it (if you can).

That is because of the initial and final power you used.

The Meg - Frame 8810

[...]

HSTM On (initial power - clipping point - final power) - 10-40-10 / 20-40-10 / 30-40-10 / 40-40-40 / 50-40-50 / 60-40-60 / 70-40-70


10-40-10 is too much here IMO.
It's interesting, in The Meg I prefer 20-40-20 much over e.g. 60-40-60, but with Black Hawk Down, I prefer 60-40-60. But of course we should pick settings which look good in the majority of situations.

Edit: I just double checked and I get completely different results for The Meg here. With 70-40-70 the face is brighter for me compared to 20-40-20, while in your images it's the other way round. Can you double check?

Edit2: Try comparing strength 100 with 30-40-80 with a strength of 200 using 20-40-20 with The Meg. Looks almost the same to me. What do you think?

Something has changed regarding color between build 87 with the 90 series. Color has noticably de saturated and the hue has changed. I had to click hue 4 points toward red to get close to build 87s default.
Hue should not have changed. Not sure why that would happen. We'll revisit saturation in the near future.

From what I read in the different forums around the world, most of people get a WOW effect with preserved saturation more than they do with preserved highlights details.
And I believe that despite Neo-xp and madshi who always favors highlight details, most people prefer saturation instead.

So let's keep that in mind when choosing the default desaturation settings.

Also since there are clearly strong different taste about that, it makes a lot of sense to propose a settings with maybe 3 strength: balanced, saturation lovers, highlights details lover.
As I said before, we'll revisit saturation later.

Experiment with the clipping point:

HSTM On (initial power - clipping point - final power) - 20-10-20 / 20-20-20 / 20-30-20 / 20-40-20 / 20-50-20 / 20-60-20 / 20-70-20 / 20-80-20

20-40-20 is the sweet spot here.
Finding the best settings might be difficult because it seems to differ a lot depending on which image you're looking at. Also, several of these settings also have an indirect influence on the overall strength of the algorithm. E.g. smoothing decreases the strength to some extent. I suppose changing the clipping point or power values will probably also affect the overall strength somewhat. So is it better to pick clipping point / power values which have a lower algorithm strength overall, and then to increase the strength to 200? Or maybe we should decrease strength to 50 and pick clipping point / power values which raise the strength of the algo?

JFYI, the two power values aren't directly related, so it's perfectly valid (in theory) to pick different values for both.

All that said, 20-40-20 is what I settled on after a very quick and rough test with various test scenes/videos. But I'm far from sure that these are the best settings for the widest range of content.
 

·
Registered
Joined
·
1,479 Posts
Discussion Starter #6,635 (Edited)
I wouldn't want you to make decisions on HSTM settings based on saturation.

Maybe I should completely disable any desaturation for now? It will look bad, IMHO, but it should make it easier for you guys to judge which HSTM settings are better and which are worse, because saturation no longer plays a role? Or maybe, it could actually go into the other direction that you pick completely different HSTM settings because you think the saturation is otherwise too high. It's difficult... <sigh>
Yes that might be a good idea. Right now I can't really like v97 compared to v96, or htsm off because of loosing desaturation.

Also, I think I you remove desaturation, it will slightly lessen the overall contrast gain of HSTM because we will get less lightness nuances. So if we still like it with no desaturation, it will be better even later.

What's important is apple to apple saturation comparaison.
So either you disable desaturation completely both with HSTM on and off.
Or maybe, you could disable any desaturation BELOW the bt2390 knee we would get with HSTM off. Right now, if I understand correctly you do not desaturate anything below the Bt2390 knee with HSTM off, but you do with HSTM ON for anything you compress below the "old" knee. So at least making that apple to apple would help. Maybe it's even "more correct" anyway.
 

·
Registered
Joined
·
1,208 Posts
It's interesting, in The Meg I prefer 20-40-20 much over e.g. 60-40-60, but with Black Hawk Down, I prefer 60-40-60. But of course we should pick settings which look good in the majority of situations.

Edit: I just double checked and I get completely different results for The Meg here. With 70-40-70 the face is brighter for me compared to 20-40-20, while in your images it's the other way round. Can you double check?
That is probably due to the difference of the DT value and real nits.

With [email protected]:

HSTM On (initial power - clipping point - final power) - 20-40-20 / 70-40-70




With [email protected]:

HSTM On (initial power - clipping point - final power) - 20-40-20 / 70-40-70



20-40-20 is my favorite for both, but with [email protected] you can see the difference in depth much better.

Edit2: Try comparing strength 100 with 30-40-80 with a strength of 200 using 20-40-20 with The Meg. Looks almost the same to me. What do you think?

With [email protected]:

HSTM On (strength - initial power - clipping point - final power) - 100-30-40-80 / 200-20-40-20




With [email protected]:

HSTM On (strength - initial power - clipping point - final power) - 100-30-40-80 / 200-20-40-20



200-20-40-20 is better than 100-20-40-20 here, I guess? I don't know for sure...

I didn't play with the strength of the HSTM algo yet (or with different values for initial and final power, because the result seemed more balanced with both powers at the same value).

I will test different strength values with this scene and others too see if there is a sweet spot for it too.
 

·
Registered
Joined
·
410 Posts
Still doing testing...

but found one issue with the new HSTM algo....Atomic Blonde (especially the second Atomic Blonde image with the close-up)

I'm testing the following atm


htsm off and sky off
vs
htsm off and sky on
vs
htsm on (at default) and sky off
vs
htsm on (at default) and sky on (at default)


First image:


Lost detail in the hair with:

htsm on (at default) and sky off to some extent;

and certainly so with htsm on (at default) and sky on (at default) when compared to the option being off







and even more so in this image, where there's lost detail in the hair with both:

htsm on (at default) and sky off;
and htsm on (at default) and sky on (at default).









Also, just one other


I looked at this Harry Potter scene

htsm on and sky off hist smoothing 0
vs
htsm on and sky off hist smoothing 1
vs
htsm on and sky off hist smoothing 20
vs
htsm on and sky on hist smoothing 0
vs
htsm on and sky on hist smoothing 1
vs
htsm on and sky on hist smoothing 20


Depending on the combination (with largely default settings), there can be some issues with the beam, which repair lum channel was intended to fix, there's less luminance in the centre compared to the edges (most noticeable on right side of image). This might be fine tuned with other settings but I haven't done that yet

 

·
Registered
Joined
·
1,208 Posts
Still doing testing...

but found one issue with the new HSTM algo....Atomic Blonde (especially the second Atomic Blonde image with the close-up)
Hi Fer15,

I tested myself, and I also have this issue, even with a strength of 1 :eek:

Hopefully, it is just a bad color gamut conversion or something easily fixable.

Depending on the combination (with largely default settings), there can be some issues with the beam, which repair lum channel was intended to fix, there's less luminance in the centre compared to the edges (most noticeable on right side of image). This might be fine tuned with other settings but I haven't done that yet
That looks bad, but isn't it just a problem of balance between saturation and luminance that will need to be re-adapted for the HSTM algo (saturation boost 1 or 2, measured desaturation, or something like that)?

Should we take this into account for now?
 

·
Registered
Joined
·
410 Posts
Hi Fer15,

I tested myself, and I also have this issue, even with a strength of 1 :eek:

Hopefully, it is just a bad color gamut conversion or something easily fixable.



That looks bad, but isn't it just a problem of balance between saturation and luminance that will need to be re-adapted for the HSTM algo (saturation boost 1 or 2, measured desaturation, or something like that)?

Should we take this into account for now?

I think only Madshi can answer both of those questions :D


Hopefully both issues can be easily fixed.
 

·
Premium Member
Joined
·
727 Posts
As I mentioned before, saturation is a topic we will revisit later. I wouldn't want you to make decisions on HSTM settings based on saturation.

Maybe I should completely disable any desaturation for now? It will look bad, IMHO, but it should make it easier for you guys to judge which HSTM settings are better and which are worse, because saturation no longer plays a role? Or maybe, it could actually go into the other direction that you pick completely different HSTM settings because you think the saturation is otherwise too high. It's difficult... <sigh>.

It is more difficult than I anticipated. Thank you for the response and sorry for the saturation derailment.
 
6621 - 6640 of 9410 Posts
Top