MadVR in use with LG OLED Thread

[QBhd]

The S&M Demo does come in all kinds of flavors... there is a 600 MDL, 1000, 2000, 4000 , and 10000... Plus various colour spaces too... and a few other differences. There is 13 different versions.

But here is the thing... If it is not needed, more DTN will do nothing for us. So the lower nit sources will not need as much. I went through ALL 71 scenes, the handful of scenes where a higher DTN than 65 helped is not worth the more than 1/2 of the scenes where it got dimmer.

QB

[quietvoid]

I mean profiles would be great by MaxCLL, but it would probably be a bit painful on maintenance and keeping them updated.
I'd probably do 75/1000, 100/2000, 125/4000, 150/10000 or something.

[chros]

150 is not too much at all.
While some scenes of the demo benefit from some higher brightness with setting it lower, others lose contrast and highlight detail.
The one I noticed was the waves hitting the rocks, there's detail lost with DTN at 50.
What about 65/75?
If it is not needed, more DTN will do nothing for us. So the lower nit sources will not need as much. I went through ALL 71 scenes, the handful of scenes where a higher DTN than 65 helped is not worth the more than 1/2 of the scenes where it got dimmer.
There's a big difference between 65/75 vs 150! :) (Note, I haven't gone through it so detailed as QBhd did.)

But here are couple of things that I don't understand:
1. DTN should be only (!) responsible of taking (high) FALL into account!
- it doesn't do anything with frames that their peak is below DPL: so if you use 1000 DPL then it doesn't do anything with frame peak below 1000 nits
- when DTN is disabled, it should not affect content with "normal" FALL, madvr's other algo should take care of it
2. high fall has nothing to do with maxCLL
- although you can argue that low maxCLL (~700) titles definitely won't get high FALL
- take a look at this list (https://docs.google.com/spreadsheets/d/1-t8_mF-jxTbArQIvNHm7XQUcMFUTvJ5O87bLFS1lCm0/edit#gid=1677456743), order by avgFall
- SM 10000 HDR10 has these values: MaxFALL: 1667, AvgFALL: 311, AvgFMLL: 3576 nits
3. if you still (!) feel the need of profiles based on whatever attribute, then it only means one thing:
- the algo that madvr uses for DTN with HDR passthrough is wrong!
- in this case you have to ask madshi to modify it according to your needs

[QBhd]

I see zero need for profiles with regards to DTN

QB

[SamuriHL]

Until we take a look at if they can truly help or not, I'm not ruling them out. If you have no need for them, that's great and it simplifies your configuration.

[SamuriHL]

Build 115 is interesting. That new option to not apply peak nit to DTN does some crazy stuff for me at least. I can't get the brightness to come back down to something reasonable no matter what I set the DTN to after enabling that option. I had to disable it. After playing around and a fair bit of testing tonight, I'm back to using DTN 150. I'm also using desat i + 75% dumb as it gives just a slight edge of saturation over desat i 1+2 + 50% dumb. desat i 1+2 + 50% dumb certainly seems correct technically speaking from what I can see. But for now I'm running desat i + 75% dumb just to play around with it. Hopefully someone can tell me what the hell the do not apply peak is supposed to do. I don't know if it's something we should care about or not but so far for me, not is the answer.

[Magik Mark]

Guys Can you share ur new setting under b115. Thanks a lot

Sent from my SM-G988B using Tapatalk

[quietvoid]

I've played with the new DTN option a bit but at a certain point it stops changing the still image (same as just changing DTN sometimes), gotta test more :)
I think we're supposed to be comparing the target tonemap nits on the OSD.
Oh well, looks like Neo-XP is testing too :)

Guys Can you share ur new setting under b115. Thanks a lot
There's not much that changed, probably just have to reset the saturation option (mine defaulted to y' after changing to 115).

[SamuriHL]

As mentioned, I'm using Neo-XP's desat i + 75% dumb and I like it so far. it gives just a tad bit more saturation than desat i + apply 2 + 50% dumb which is good for us IMO. YMMV of course. Also I had to up the DTN to 150 even with this new do not apply peak nits setting disabled. I need to play with that some more. I don't think that should have changed without the option enabled. I can tell you with that option enabled nothing I did in the DTN setting made the brightness even remotely ok. It was absurdly bright for me.

[aron7awol]

I don't actually see any saturation issues on our high-nit displays. I prefer desat i 1 only, no dumb. Is there some specific content that you guys have seen saturation issues with, even with our ~1000 nit DPLs?

I did some extensive testing with BvS and I definitely preferred NOT checking "disable avghighlights ceiling". I find the ceiling is a good thing, as it prevents DTN from kicking in too strongly. I really liked this combined with 75 DTN.

DPL 1000
DTN 75
unchecked "disable avshighlights ceiling"
desat i 1 only
Pretty much everything else disabled.

[chros]

Hey Aron, welcome here!
What about the new killer feature "don't add DPL" ? :) That's the biggest change for us since DTN was invented!

[SamuriHL]

Are you able to use it without overblowing the brightness?

Sent from my SM-G975U using Tapatalk

[aron7awol]

I haven't tried the "don't add DPL" feature yet. It sounds like maybe there is a bug and it isn't working right, but it's not really a big deal. It mostly just makes it a bit more consistent to dial in DTN across different DPLs, and if you guys settled on using a particular DTN before, you'll probably have to experiment to find your new DTN setting with that box checked. But in my testing I really liked the avghighlights ceiling kicking in when it did. It kind of makes DTN a bit less important if DTN is set high enough to hit that ceiling in a lot of cases, but in scenes that don't hit the ceiling there are still noticeable differences between different DTN settings, especially lower ones like 50 vs 75.

Trying to figure out what is highlights and putting a ceiling on the TM target always made sense to me in principle, especially on high-nit displays. It essentially avoids unnecessarily dimming the overall picture for highlights only, but still will dim the picture if the whole scene is really bright and it's not just highlights, and in that scenario we need to dim the picture to fit it all in with good detail. This results in a much more consistent experience as far as brightness, especially on a high-nit display that can really handle 95% of the scenes out there without having to dim the overall picture, and only when we have scenes that the whole scene is pushing the limits do we need to dim everything (which ABL would likely do anyway).

[chros]

I agree, and that's why the new feature that you and neo-xp asked for does the same thing.
Check that box and set 60 DTN (I used 55 without it previously) and try out different scenes/movies/tv shows. Don't forget that with high DPL (e.g. 1000) DTN is not used at all if the measured peak is below the set DTN.
The result is just amazing! No more unnecessary dimming with low FALL titles!
My opinion is that this is how it should have worked from the beginning! And that means we watched the wrong images in the last 1.5 years! :)

And here's a riddle for you guys: try to figure out what formula the DTN "algo" uses :D
But seriously, ask madshi about it, because it's insane that we have to fool around something to try to figure out how it should work...

[SamuriHL]

If any of you have the SM disc, here's what I want you to do.....set your favorite settings, whatever you want. Now, play the 1000 nit version of the demo file and go to 6:04. You'll see a little croc on some rocks. To the upper left there's some yellow and black mixed in with the rocks. Now check it out in the 10000 nit version. Do they look the same?

[aron7awol]

Well, I just felt that eliminating DPL from the calculation made for more consistent DTN results across different DPLs. I made a spreadsheet when we were discussing it so I could compare the results with both formulas. Even with it eliminated, I still feel the avgHL ceiling is beneficial, as it adds some additional intelligence as far as when to dim the whole image.

I'm glad to hear it seems to be working for you though, maybe there is no bug after all. :)

madshi alluded to there being a slight tweak to the formula, but didn't share it, so I got the feeling he didn't want to for whatever reason.

After my recent testing and posts in the AVS thread, I'm also thinking don't desat might actually be a reasonable option for us TMing to 1000 nits.

[chros]

We don't know whether there's a bug or not because we don't know how it's supposed to work :)
Maybe asking him via PM is more fruitful.
I think desat2 is needed even for us, again, these options don't do much when frame peak is below DPL, but when it's above then we have the same issue as the projector guys.

avgHL ceiling is interesting, never played with that one. Here's a bit historical overview:
- it was created by soulnight along with the FALL algo (DTN) for pj users
- madshi modified the original DTN algo and stated that this option is maybe not needed anymore
I think this option works the most when there's high FALL so "higlights" are not even. But when we got only couple of small light source it doesn't do much.
But again, we don't know how he defined "highlights"... :)

@SamuriHL, I don't have the 1000 version, if you can upload it then I can check.

[aron7awol]

Well I think we know how it's supposed to work close enough to know if there's a bug or not. The difference between what we calculate based on the "known" DTN formula and what madVR reports has been very close.

I'm not referring to frame peaks below DPL, of course saturation is irrelevant in those cases as it doesn't tone map. But the scenes I tested in BvS and MMFR all had peaks well above DPL, and less desaturation looked better. Then I tested MMFR with the LG TM and madVR with the different desat options and don't desat was closest to LG's approach, and it not only looked better but also matches the non-TM version best. Now maybe a small amount of saturation (less than 100% desat 1) would be the best balance, I'll have to test more.

I believe highlights are defined as everything >100 nits for the avgHL calcs.

[chros]

"highlights are defined as everything >100 nits for the avgHL calcs"

I hope not (don't think so), because when the whole content is shifted (like The Meg, SM 10000 demo, etc) it wouldn't work. That's what I meant how he defined it.
But maybe that option is useful indeed, I also have to play with it and see, but with normal content and not with demos :)

About desat2, you saw the discussion over there, it's not that simple. I think James is right, fire should be yellow not the nonTM version!
And I wonder whether your CX acts the same with TM as our 2018 models :)

[aron7awol]

I did some testing with The Meg, and I actually preferred it with the ceiling in most shots. Obviously this stuff comes down to personal preference, but I actually am really liking when it is kicking in, it makes the image look more realistic to me, while without the ceiling it dims too much and no longer looks like an outdoor scene. Or as another example inside the elevator, with the ceiling there is a really good balance between the brightness of the light and the brightness of the walls and people, it looks realistic, but without the ceiling, it dims and the light is still really bright but everything else is too dim relative to it. FWIW, I did a bunch of A/B comparisons for my wife from The Meg, who has no idea what I was even doing other than just asking her which she preferred, and she chose the ceiling every time. The same happened earlier today when I did the same with some stuff from BvS. So all in all, I'm liking it and I actually think it was tunable rather than just 2xavgHL it could be really great and useful even for people that might prefer a bit dimmer than me but could still take advantage of the ceiling. I have similar feelings as you just described with the DTN algo tweak: no more unnecessary dimming!

Yeah, the desat/hue stuff is a whole other can of worms, and I actually think that the choices for TMing to 1000 nits or 100 nits maybe should be drastically different. It's almost as if the hue shift helps offset the flatness of the heavy compression when going down to 100 nits, which does seem potentially desirable. But for us and TMing to 1000 nits, I actually preferred the no or very little desat approach, and I think having ~1000 nits to throw at it is the difference there. Because while everything looks pretty flat in my terrible photos in that thread, having that fire with ~1000 nits it looks totally not flat and lively and realistic. FWIW, my Sony does shift some of the fire to yellow even with its 1000 nits, and it doesn't look as good to me when comparing side by side. Since I got the CX I've had it and the Sony side by side on dual outputs from my AVR so I've been able to compare a ton of content that way, which has been great.

We just watched Ghostbusters (2016) with the avgHL ceiling and don't desat and the results were phenomenal. But obviously without a frame of reference that's not worth much, so after the movie was over I went back to a whole bunch of the brightest frames and compared with/without the ceiling, and separately with/without desat. I once again preferred the ceiling and don't desat, and this has become my normal setting.

[SamuriHL]

I can't upload that demo video unfortunately. But that disc is worth having for this very reason. Looking at it at 1000 gives us some sense of what it should look like when tone mapped from 10000 -> 1000. Right now for testing purposes I have all the enhancements off, ceiling enabled (I agree, that's an interesting thing to play with....I've always had it off), DTN 65 and I forget what my desat is set to. most likely desat i + apply 2 at the moment. That scene at 6:04 looks very wrong to me. I can't tell but it almost looks like it's shifting the white to almost pink in some parts and it's WAY overexposed and loses the detail. Now I turned highlight recovery, shadow recovery, etc all off right now but even with some of those on I could NOT get madvr to tone map that to look correct. Which is why I want to see if anyone else can and what settings they use. This is very much a torture test I know. But it's indicative of the problems we face when even doing 4000 to 1000 tone mapping.

[aron7awol]

I'll take another look at the S&M disc with my new settings and report back.

On the highlights ceiling, even in a case of a scene from The Meg with really high FALL, avgHL will always be equal or higher than FALL since it is basically FALL but without the <100 nit pixels. And so the ceiling ends up as double that which is still quite high on those high FALL scenes. Now maybe 2x isn't quite enough, I'd love to have to configurable.

[chros]

I can't upload that demo video unfortunately. ... Looking at it at 1000 gives us some sense of what it should look like when tone mapped from 10000 -> 1000. Right now for testing purposes I have all the enhancements off, ceiling enabled (I agree, that's an interesting thing to play with....I've always had it off), DTN 65 and I forget what my desat is set to. most likely desat i + apply 2 at the moment. That scene at 6:04 looks very wrong to me. I can't tell but it almost looks like it's shifting the white to almost pink in some parts and it's WAY overexposed and loses the detail. Now I turned highlight recovery, shadow recovery, etc all off right now but even with some of those on I could NOT get madvr to tone map that to look correct. Which is why I want to see if anyone else can and what settings they use. This is very much a torture test I know. But it's indicative of the problems we face when even doing 4000 to 1000 tone mapping.
This is my b115 settings (https://www.avsforum.com/threads/improving-madvr-hdr-to-sdr-mapping-for-projector.2954506/page-438#post-59228954): I also disabled contrast recovery as well, sometimes it ruins it :)
So basically almost all the options are disabled due to bugs in them or just don't work at all. :)

And I disagree with the validity of the 10000 vs 1000 test as I said above, due to this (https://www.avsforum.com/threads/improving-madvr-hdr-to-sdr-mapping-for-projector.2954506/post-60365341).
And what madshi said when he was back 1 month ago, what do we want from madvr to do: keep the nonTM image or fix the "broken" master? :) In this case (mad max explosion(s)) the latter :)

On the highlights ceiling, even in a case of a scene from The Meg with really high FALL, avgHL will always be equal or higher than FALL since it is basically FALL but without the <100 nit pixels. And so the ceiling ends up as double that which is still quite high on those high FALL scenes. Now maybe 2x isn't quite enough, I'd love to have to configurable.
You convinced me, I also disabled the "disable avg HL" (as you can see above), this is when I played with it (https://forum.doom9.org/showpost.php?p=1930893&postcount=1208) a bit.

I made a spreadsheet when we were discussing it so I could compare the results with both formulas. Even with it eliminated, I still feel the avgHL ceiling is beneficial
I'd be interested in it if you can share it (e.g. via google spreadsheet), here's my simple test result with "don't add peak nits":

FALL DTN to reach >700 DPL
15 -
24.27 -
24.58 500
24.75 496
27 457
33 369
42 295
49 249
67 184
76 162
100 123
130 95
161 77
207 60
245 51
286 43
353 35
449 28
730 17
842 15
900 14



FALL DTN to reach >800 DPL
27.48 -
27.7 497
28 491
220 63
231.64 60
237 59

[SamuriHL]

The difference in this case on the 10000 vs 1000 is that not only do we have it in hdr we also have it in dv. And it's not broken in dv. Madvr is doing something whacked here.

Sent from my SM-G975U using Tapatalk

[chros]

Hmm, doest it change with the differernt desat options?

[SamuriHL]

I need to test all of them but the ones I tried...no, not really. None of them fixed the issue.

[SamuriHL]

Ok, this is definitely a problem. The issue goes away when I enable compromise ON. WTF? This is messed up. Whatever madvr is doing in the hue shift with compromise off is screwing this up. I even tried desat i and nothing else enabled and it's still wrong.

[chros]

Cool, tell madshi, post pictures etc,...

And if you guys are womdering why I tell You guys to do it:
none of my 2 recent bugs were replied (by anyone, but you, when I asked You guys), let alone fixed. Hence I stopped participating there.
If you still be interested in those and other possible future observation of. mine, I can post them here, but You have to force to be fixed :)
I'm tired of this ....

[SamuriHL]

I hear you. There's a lot of...."stuff" going on lately that isn't awesome. I finally fixed the ability to grab screen shots so I will go post them.

[chros]

Cheers, we can still work together here for the greater good if that's what we want... :)

[SamuriHL]

I think we are going to have to if we want to drive these OLED panels to their best selves. LOL We're constantly learning and they're constantly changing things that require us to retest our previous assumptions.

[aron7awol]

I just posted in the AVS thread, but after some additional testing, I'm finding max to be SO much better than sep. It fixed all the issues I was having with too much desat, even with just desat i 1. I think that combo is perfect for us with OLEDs.

[SamuriHL]

Yea I saw some people playing with that. I've not had a chance to test max yet. Will give that a go tonight.

[chros]

I just posted in the AVS thread, but after some additional testing, I'm finding max to be SO much better than sep. It fixed all the issues I was having with too much desat, even with just desat i 1. I think that combo is perfect for us with OLEDs.

Cheers, but take a look at the mmfr sample from 0:15-0:30 : top left, big reflection on the edge of the car.
With max+desat-i+nodumb (no desat2): those 2 big patches become blue! Nothing helps it.
This is way more annoying than some color shift.

Also at 2:15 the sundown on the left, becomes orange (but we know that it's probably the source.)

NeoXp reported issues with max couple of weeks ago and if I have to bet, nothing happened to it.

[aron7awol]

I don't have the sample clips. Do you know the timestamps from the movie?

FWIW, linear 54 was almost identical to max. Does that have the same issues?

[chros]

How come you don't have the samples?! :D

It's in the beginning, 00:01:20.
And yes, 54 is similar to max.

[aron7awol]

How come you don't have the samples?! :D

It's in the beginning, 00:01:20.
And yes, 54 is similar to max.
I didn't see where they were posted/hosted.

[aron7awol]

Ok, I see the reflection in the car, and I see that it is definitely blue with max, but it is also definitely very blue in the source! This is what I found in my other testing: max is very true to the source. And FWIW, it is sep that can create artifacts in the green spear for me while max looks true to the source there as well.

[SamuriHL]

My disc just came so I am watching it now with these settings. max, desat i, highlight 1000, DTN 65. Everything else shut off. I've only ever seen this in DV streaming so...

[SamuriHL]

I forgot what a messed up grade this is (I've not watched it in a while and as mentioned it was DV streaming). These settings seem quite faithful to the grade. You aren't going to fix the source with these settings, but IMO it's not for us to fix. I'm rather impressed.

[aron7awol]

I can't upload that demo video unfortunately. But that disc is worth having for this very reason. Looking at it at 1000 gives us some sense of what it should look like when tone mapped from 10000 -> 1000. Right now for testing purposes I have all the enhancements off, ceiling enabled (I agree, that's an interesting thing to play with....I've always had it off), DTN 65 and I forget what my desat is set to. most likely desat i + apply 2 at the moment. That scene at 6:04 looks very wrong to me. I can't tell but it almost looks like it's shifting the white to almost pink in some parts and it's WAY overexposed and loses the detail. Now I turned highlight recovery, shadow recovery, etc all off right now but even with some of those on I could NOT get madvr to tone map that to look correct. Which is why I want to see if anyone else can and what settings they use. This is very much a torture test I know. But it's indicative of the problems we face when even doing 4000 to 1000 tone mapping.
So I just tested this scene. This is one where the 2xavgHL ceiling is not high enough, and it really needs to be 2.5x or 3x. Then it looks quite good to me and very similar to the 1000 nit grade.

[aron7awol]

Yeah, MMFR really has quite a jacked up grade! But I agree, if we start messing with our settings to make it look more "normal" we'll just end up screwing with everything else. So I'm with you, it's not for us to fix, let it be what it is.

[SamuriHL]

So I just tested this scene. This is one where the 2xavgHL ceiling is not high enough, and it really needs to be 2.5x or 3x. Then it looks quite good to me and very similar to the 1000 nit grade.

That would explain it. Good. Maybe madshi will consider letting us set a scale for the avgHL ceiling. 3x would probably cover everything.

[SamuriHL]

Yeah, MMFR really has quite a jacked up grade! But I agree, if we start messing with our settings to make it look more "normal" we'll just end up screwing with everything else. So I'm with you, it's not for us to fix, let it be what it is.

I realize this is a huge philosophical discussion but I fall in the madshi camp on this one. We want to reproduce what's on the disc, not what "looks good". If people want to chase that elusive "fix the source" viewing, have at it but allow us who want to remain faithful to what's on the disc to do so. If the person who grades the movie has no f***ing clue what they're doing or using s****y equipment to grade on, so be it. LOL I want accuracy.

[chros]

:D OK, so we all agree that mmfr is screwed up. There's indeed a small blue patch inside of those reflections at 10000 DPL.
But this is interesting, it means we can use the following to correct bad master (e.g. manually via profile): sep+desat-i+nodumb+desat2

OK, I'll switch to max+desat-i+nodumb , and will create a profile for the above.
@Aron, does the highlight setting (0-1000) matter in this case? (And do you know about profiles in madvr?)

And I linked the samples here (https://www.avsforum.com/threads/improving-madvr-hdr-to-sdr-mapping-for-projector.2954506/post-60322083) for you.

[aron7awol]

Yeah, I don't see a problem with a profile for a specific really strange of bad grade like MMFR. sep instead of max, maybe fire tweak, whatever else makes sense. After all, what looks best is the #1 priority for me always. But I just don't want to adjust my daily driver settings because of a bad master, which I'm sure you guys feel the same way on.

I didn't notice any difference from the highlight setting in some quick testing, but I didn't do too much so I'm sure there are some instances it has an effect? If you guys know of any or find any, please share.

Yes, I use profiles all the time to switch back and forth for comparisons. That's why when I saw the guys on AVS trying to compare the desat stuff and using side-by-side different displays and all that, I was wondering why they wouldn't use HDR output and just toggle between profiles via hotkey. It's by far my favorite way to compare all of this stuff.

Thanks for the link to the samples!

[SamuriHL]

I see that madshi agreed to your avgHL multiplier option. That should be fun to play with. I had the highlight set to 1000 for MMFR last night and didn't seem to make a huge difference, but, I was watching the movie rather than pixel peeking so it's possible it had an impact. But if so it wasn't glaringly bad.

It's a lofty goal to try to create profiles to fix grading/source issues to make them look better. The only time I'd really want to go there is if the grade is so egregiously bad that it completely pulls you out of the movie. MMFR, as whacked as the grade is, seems fairly intentional. Atomic Blonde is another crazy grade which is very much intentional. Going around "fixing" these....I don't know, just not my thing I guess.

[chros]

@Aron "What is the actual formula?"
:D Thanks, that's why I said couple of days ago that I pay a drink if someone can figure this out :D
Also said, that when he implemented soulnight's DTN formula, he changed it, only god and he knows to what :)

And that's why I chose 60 DTN, if it would be higher than it dims too much with very high FALL.

[aron7awol]

Reverse engineering FTW! :)

I'm confident I figured out the formula:

log10(DPL)*DTN/50*FALL and optionally +DPL

[chros]

:D That's what I meant!
But do you think that DPL is still in there???