MadVR in use with LG OLED Thread [Archive] - Page 32

quietvoid

20th February 2021, 17:20

I wouldn't say that frame at 22s is unnecessarily dimmer, some of these lights are noticeably closer to being blown out with the default curves.
You can just make it brighter with dynamic clipping at 50 (for this specific scene), without affecting the detail.

This is also so close that I'm not sure you'd appreciate the difference in HDR playback.

I don't believe (anymore) that curves should be used to increase brightness, because you're definitely going to lose detail that way.

SamuriHL

20th February 2021, 17:21

I tend to agree. Use the curve to retain as much detail as possible. Brightness vs detail is the compromise, as madshi has said numerous times. You give up one to have the other as they are on opposite sides of the spectrum. TV algorithms it's pretty clear to see the choice they make, unfortunately.

chros

20th February 2021, 20:03

You can just make it brighter with dynamic clipping at 50 (for this specific scene), without affecting the detail.
This was a really good idea, thanks, I updated the post over avsforum:
"dynamic clipping has still an effect on the resulting curves/image (tested with MDL 1000, frame peak before dc 1621 and after dc 1186, so it's above MDL), so it clearly doesn't clip at MDL."

I wouldn't say that frame at 22s is unnecessarily dimmer, some of these lights are noticeably closer to being blown out with the default curves.
"Closer" means they are still not, so why degrade the image? :)

This is also so close that I'm not sure you'd appreciate the difference in HDR playback.
That's true but this can be said about all the other options as well, even with sources that we know, let alone those that we watch for the first time :)

I don't believe (anymore) that curves should be used to increase brightness, because you're definitely going to lose detail that way.
At some point, yes, as we can see with the meg and SM, but the question remains: which is that point? :)

The curves are not meant to fix the "issues" in MMFR, that's a saturation problem.

Err, I think these curves are supposed to fix the saturation issues at the first place, although MMFR is an extreme example.

SamuriHL

20th February 2021, 21:32

No, madshi said saturation will be revisited after the curves are selected. There are people who are choosing to have desaturation options still enabled for some content so the curves are not going to fully fix saturation. They will have an impact, certainly.

aron7awol

21st February 2021, 06:06

So, as far as I understand how this works (please correct me if I'm wrong):
- DTN with curves only selects (interpolates between) the right curve(s):
Agreed, and I guess I'd say this is sort of all DTN ever did, was set the BT.2390 curve used based on peak and the resulting compression of target/DPL.
-- if DTN is disabled with curves, only the x1 curves are used
Agreed, as without DTN there is no compression.
- with 800/870 DPL, only the x1/x2 curves will be used with 2230 (neither the 4x/8x/16x ones), because x2 means 1740 ADPL with 2230 which is really high already
Well, a frame with peak and target >1740 would result in compression >2x so it should interpolate using the 2230x4 curve as well in those scenarios. And then even as we go further to frames and targets >3480 it would use that 2230x4 to interpolate along with 2230x8 (and 10000x4 and 10000x8), and even more extreme cases with frames > 6,960 we would interpolate between 2230x8, 2230x16, 10000x8, 10000x16. So really, almost all curves can be used with almost all DPLs in the right scenarios.
- the frame above uses 2 curves: 509x1 and 2230x1 (but closer to 2230x1)
Agreed, since DTN doesn't kick in, there is no compression and so it only needs to interpolate between 2 curves. As soon as DTN kicks in (and thus results in compression) we interpolate between 4 curves, except in the coincidental cases where peak and/or compression happen to land exactly on one of the options. :)

chros

21st February 2021, 13:04

Well, a frame with peak and target >1740 would result in compression >2x so it should interpolate using the 2230x4 curve as well in those scenarios. And then even as we go further to frames and targets >3480 it would use that 2230x4 to interpolate along with 2230x8 (and 10000x4 and 10000x8), and even more extreme cases with frames > 6,960 we would interpolate between 2230x8, 2230x16, 10000x8, 10000x16. So really, almost all curves can be used with almost all DPLs in the right scenarios.
Yes, you are right, sorry, I meant titles under that 2230 curve.

Agreed, since DTN doesn't kick in, there is no compression and so it only needs to interpolate between 2 curves. As soon as DTN kicks in (and thus results in compression) we interpolate between 4 curves, except in the coincidental cases where peak and/or compression happen to land exactly on one of the options. :)
About this: I wanted to test elevated black with Mehanik's (https://drive.google.com/uc?id=1bAtIc0C2GCZZVbgV-ntSKED4q-9C8L5_&export=download) black clipping pattern (01. Black level\01. black-level-v1.mp4), but the image doesn't change at all when I modify the 110x1 curve!!!
Can you guys reproduce it? (either in SDR or HDR ouptut)
Video stat: frame peak 11 nits, FALL 0.051.
Curve 110x1 mod: I moved up the first 11 (!) points to around 10 nits!
There's no change in the image at all when you switch off/on the curves.
What's happening here?

Edit: it doesn't change either if you set all the points to the max (e.g. by clicking the "create bt.2390 curve now" with its default settings).

Lastly, as he stated, madvr now should clip at MDL straight away (although it doesn't seem to work now).
But if that will be the default behavior then there won't be need for DTN at all for MDL 1000 titles!!! (which I guess the 70% percent of all the content today, at least)

quietvoid

21st February 2021, 17:29

Lastly, as he stated, madvr now should clip at MDL straight away (although it doesn't seem to work now).

I had posted an example where the MDL clipping worked: https://www.avsforum.com/threads/improving-madvr-hdr-to-sdr-mapping-for-projector.2954506/page-568#post-60401221

quietvoid

21st February 2021, 20:13

Anyways I ended up making a minimum steepness + half max knee file, seems to be better than what I had done for Neo-XP's adapted one.
It's in my signature.

Alternatively, here's Neo-XP's 1.2 curves adjusted for 870 DPL: https://gist.github.com/quietvoid/d797c47827d39651822d21aa56b77beb
It's a bit more contrasty/darker, see https://www.avsforum.com/threads/improving-madvr-hdr-to-sdr-mapping-for-projector.2954506/page-607#post-60518938

Since it lowers brightness a bit more, I lowered DTN and increased the FALL detail to gain back some brightness. But again that's totally subjective.
Settings: https://postimg.cc/n9Fr9s7r

Now I need to start actually watching stuff.

chros

21st February 2021, 20:28

I had posted an example where the MDL clipping worked: https://www.avsforum.com/threads/improving-madvr-hdr-to-sdr-mapping-for-projector.2954506/page-568#post-60401221
Omg, that was a month ago and I almost forgot about it :) Thanks
Yes, that option works in b123 but not with b124/125.

Anyway, can you guys experiment with that option at the MDL value if you will have time whether it's needed or not? I'll do the same.
Because if he fixes it there won't be turning back ... :)

Anyways I ended up making a minimum steepness file, seems to be better than what I had done for Neo-XP's adapted one.
It's in my signature.
Cheers, I'll take a look some other day.

Now I need to start actually watching stuff.
:D Sure

PS: I gathered up the madvr beta changlog (https://pastebin.com/6N9tr0t6) into 1 file from b100 (I haven't gone back more, yet?), I'll plan to update it in the future (I've also put the link in the post of the link in my signature).

jigar24

26th February 2021, 05:58

My 1080p bluray player (couple years old) has options for 4:2:2 ycbcr, 4:4:4 ycbcr, or RGB. With the UHD players I believe they offer those choices too, along with up to 12bit output. By default on 'auto' pretty sure they all output ycbcr at either 4:2:2 or 4:4:4

quietvoid

7th March 2021, 07:16

chros

7th March 2021, 11:24

Not too sure how much the new normalization affects brightness, I haven't compared previous versions.
This should be an option, again, you can't watch content normally to actually see what the curves do, unless you revert back to b128 (but there's the mdl clipping :) ). That's why I asked him to clarify that.

So, for all of these that happened in the latest builds, I went back to b113, until everything settles down :)

Different question: did any of you calibrated DoVi? If so, do you still have the DoVi config file that can be shared?

chros

7th March 2021, 11:52

Different question: did any of you calibrated DoVi? If so, do you still have the DoVi config file that can be shared?
@SamuriHL, have you? :) I think you have all the tools for this, haven't you? :) It would be really nice from you if you can give it a chance ... and we finally mange to get rid of the elevated black crap :)

SamuriHL

7th March 2021, 21:51

Hell no. These LG panels, especially the 18 series, are way too finicky to calibrate DoVi. As soon as you mess with the config it screws everything up in my experience. I've left HDR and DoVi alone other than SM white point @d65.

RXP

10th March 2021, 12:15

About due for an update here.

I spent way too much time testing all this curve stuff.
I've updated my settings for 130: https://postimg.cc/RJzNZd33
Pretty much just disabled FALL detail, it was too sensitive. FALL mod at 80% seems to be decent for extra brightness without losing much.

Curve is r13 now, probably my preferred for now. Hopefully the others like it too.
Not too sure how much the new normalization affects brightness, I haven't compared previous versions.

I'd reenable contrast recovery but I might wait, because I don't really seem to need it (personally, at least) with FALL mod.

Thanks for posting your curve/settings. I'm dipping my toe into the new builds and yours is the first time I've actually liked what MadVR has done on my C6 OLED. I just amended the peak brightness to 650 nits of my panel.

Edit: just realised you did 870 to stop the internal tone mapping right?

SamuriHL

10th March 2021, 13:14

It doesn't stop it but it is to choose the best curve the TV offers internally. It maps very close to eotf on our c8. And prevents the TV from clipping. You may have to experiment a bit with the c6 to see if it responds the same or needs a lower target nit.

Sent from my SM-G998U1 using Tapatalk

RXP

10th March 2021, 14:39

Excellent thanks! I just tried it on my C9 and on the jazz demo it did wonders! Even though I really like the C9's HDR performance, nice to squeeze that extra out of it.

quietvoid

10th March 2021, 15:30

chros

10th March 2021, 16:41

Thanks for keeping the spirit alive! :) I'll take a look when we will have a useable build (e. g. with an option for brightness normalisation).

SamuriHL

10th March 2021, 18:42

Excellent thanks! I just tried it on my C9 and on the jazz demo it did wonders! Even though I really like the C9's HDR performance, nice to squeeze that extra out of it.

The C9 and newer give you a few more options if you calibrate the panel. Because with those newer models, you can actually measure your peak luminance and create a curve to match on the panel. Then if you set madvr to tonemap to that measured peak, it should follow eotf to that point and wring every ounce of luminance you can drive out of that panel.

SamuriHL

10th March 2021, 18:43

So far the R13 curves remain my favorite, they are however extremely close to Fer15's new "low consistent" for everything below 10 000 nits.
R13 is still brighter in places and the difference is noticeable in HDR.

There's very little difference in highlight detail preserved, and the only I've found were for low nits (509 curves) at 130 DPL. R13 still has better contrast although it darkens shadows a little.

At this point I'm just trying to hypertune it as much as I can, and waiting on Fer15's opinion about it.

I like R13 a lot myself. I've not tested Fer's curves yet. I will be watching a movie tonight so perhaps I'll throw it on there and see what it looks like. Did you modify it for 870 nits by chance?

EDIT:

Just saw r15 and grabbed that. That's what I'll be using tonight. Thanks!

SamuriHL

10th March 2021, 18:44

Thanks for keeping the spirit alive! :) I'll take a look when we will have a useable build (e. g. with an option for brightness normalisation).

I know everyone keeps complaining about brightness normalization but everyone seems to fail to understand WHY he put it in there. It's so that people don't have to try to figure out brightness matching to compare curves. That option is not likely to be permanent once the curve(s) are selected.

quietvoid

10th March 2021, 20:06

There isn't a huge difference between R13 and R15 FWIW, it's mostly gonna be a difference for 10 000 nits and to retain contrast better instead of being flat.

SamuriHL

10th March 2021, 23:19

Well, we're at that point and that's where we want to be. :) That's seriously a good thing.

RXP

11th March 2021, 15:16

quietvoid

11th March 2021, 17:05

quietvoid, I did some more testing on my C6 with the default settings you gave. With the Spears and Munsil benchmark disk it's night and day! The detail increase is huge. Now most movies have very conservative HDR grades but good to know it's there for when they don't! It's a bigger difference than I got from doing a 3DLUT in SDR, which I thought was a game changer. For anyone that's on the fence like I was about MadVR with OLED's give it a shot, you'll be thankful you did.

Compared to the TV's tonemapping, it's definitely an improvement.

I've switched to my R16 (see signature) and changed my settings a bit: https://postimg.cc/1n14s6NB

I enabled contrast recovery because the improvement is significant even at the lowest strength.
There's slightly more blown out small highlights but the brightness is increased, and I most likely wouldn't notice the difference in highlights.

Bumped up DTN and lowered FALL highlight mod.

I'm still really conflicted with the desat lum method.
sep improves brightness and contrast significantly in S&M but I also absolutely can't stand how fires lose detail and saturation.
However it might end up being the leftover option since everyone uses it.

Hopefully the denormalization doesn't require too many changes.

SamuriHL

11th March 2021, 17:53

I was noticing the fire issue last night, as well. This may be something madman can address down the road when we get back to that. Interesting changes you've made to the settings, though. I'll check them out later, as well.

quietvoid

13th March 2021, 06:11

Another day another curve..
R17 is my most accurate curve yet, since it doesn't alter shadows much if at all.
I've compared against clipping and most of the time R16 has slightly brighter shadows, reducing contrast.

R17 is also slightly better with highlights, as long as other FALL/DTN options don't blow them out.
All this isn't very noticeable at high nits and HDR, but it's good to know that it's tracking PQ correctly.

Tweaked the settings a bit just because, I guess: https://postimg.cc/303ZkP1w

RXP

13th March 2021, 06:56

Thanks quietvoid, really excited to watch my first full HDR movie with your settings! - out of interest are you using v130 for watching content on your OLED or just testing? I flipped my OLED watching machine to v128 due to brightness norm and my PJ is on v130 as it's only 95nits peak

quietvoid

13th March 2021, 07:15

I'm just testing. I haven't watched anything in months.

chros

13th March 2021, 09:21

I'm just testing. I haven't watched anything in months.
:D Not good, not good :)

quietvoid

13th March 2021, 22:24

True.. it's even worse considering most content out there doesn't get any improvement from madVR.

quietvoid

14th March 2021, 06:45

So another curve that's mostly more for high nits, because it's even brighter for lower nits and I doubt people over at AVS are liking these much, even though they're accurate.
My idea is that a lot are used to darker video caused by having to rely on DTN for detail.

R18 is basically a brighter R17, without affecting highlights.
Since it's literally free brightness, I've gone to lower DTN and enabled max again. Settings: https://postimg.cc/c6Xs9dMT
The lower DTN is because max does reduce brightness in saturated areas, so it counteracts that loss.

max is noticeably better at highlight detail than sep, even though some specular highlights are dimmer because no desat.
I still think the significantly better fires and slightly more detailed highlights are worth losing a little contrast.

SamuriHL

14th March 2021, 15:36

I would agree with that assertion only because we have the room for it at 870 nits. Those who are trying to compress everything to a target below 100 nits have WAY less headroom to play with and thus have to make larger compromises. I'm good with losing a little bit of contrast to retain better highlights and better fire. I see you nuked FALL in your latest settings. Interesting.

quietvoid

14th March 2021, 16:13

Yea, it's not as easy for lower nits.
When tracking 1:1 with less brightness, the result is just flat because highlights aren't as bright.

I still think it's not that bad at 130 nits but not sure how it would look at 50 nits.
There's still a huge contrast increase for 2230+ nits because the original rendering intent is high contrast HDR, which 1:1 helps preserve.

As for FALL, since there's more brightness there's not much brightness difference between DTN 80 + FALL 90% vs DTN 75 + max.

I've just noticed that the 10 000 nits demo becomes flatter in some places with max because it's not desaturating, while sep keeps some contrasty bits in there.
Hard to tell how it's supposed to look, though.

SamuriHL

14th March 2021, 17:21

Compare it to the 1000 nit version. :)

quietvoid

14th March 2021, 17:48

I'm talking about the LG demo here, but it might be useful to compare S&M as well, I guess.

SamuriHL

14th March 2021, 18:14

Ah, ok. But yea, the SM demo is nice cause we have so many ranges of NIT values to compare the same content at. It can be useful.

Chouonsoku

15th March 2021, 07:09

I promise I searched before asking this but the answers I saw were a year old and I'm not sure if I'm interpreting the answers correctly. With an OLED that isn't calibrated with a 3DLUT, what is the correct setting in the calibration section? Using madVR b130 with quietvoid's settings and R18 curves. I've been using the following with the color gamut setting on my TV set to Auto and madVR outputting as HDR:

https://i.imgur.com/vjEUiKO.png

Should this be set to DCI-P3 with my TV set to Wide color gamut? And are these settings fine for SDR and HDR content?

quietvoid

15th March 2021, 14:45

DCI-P3 in madVR, TV set to Auto AFAIK.

Asked the same not long ago: https://forum.doom9.org/showthread.php?p=1933960#post1933960

I don't think there's a difference with using BT.2020, though..

VBB

15th March 2021, 19:17

It depends on which HDMI mode you use. For normal mode, you want BT.709 and the TV set to AUTO. For PC mode, because you are locked into WIDE, you want DCI-P3. The settings are only for SDR or HDR to SDR tone-mapped content.

Chouonsoku

15th March 2021, 20:16

Ah I gotcha. I don't use PC mode cus of the banding so I'll leave everything where it is. Thanks you two!

VBB

15th March 2021, 21:16

Does the gamma on your TV match madVR? I suggest you change that to 2.4, especially if you watch in a dark environment.

Chouonsoku

16th March 2021, 01:09

Does the gamma on your TV match madVR? I suggest you change that to 2.4, especially if you watch in a dark environment.

I have my SDR display modes set to BT.1886.

Asmodian

16th March 2021, 01:17

BT.1886 = 2.4 on OLEDs.

chros

16th March 2021, 08:06

Not on LG: bt1886 is slightly brighter near black than g2.4

VBB

16th March 2021, 18:16

@chros - Hmmm, that's not the case with my C7. I've done many back and forth comparisons with this pattern: https://www.avsforum.com/threads/lg-oleds-how-to-enhance-near-black-detail.2579537/post-46798617

Brightness has always been at 51, which is where 0% is truly black on my TV.

Chouonsoku

16th March 2021, 19:31

BT.1886 = 2.4 on OLEDs.If that's the case then why are both offered as Gamma options? There must be some difference. I read a long time ago that BT.1886 was better but that was when I first got my C7.

@chros - Hmmm, that's not the case with my C7. I've done many back and forth comparisons with this pattern: https://www.avsforum.com/threads/lg-oleds-how-to-enhance-near-black-detail.2579537/post-46798617

Brightness has always been at 51, which is where 0% is truly black on my TV.I thought any brightness setting above 50 began to raise black levels? Also, you're saying I should set both madVR and my TV to 2.4/BT.1886 gamma? I didn't think you were supposed to modify the setting in madVR unless your display is *actually* calibrated which mine is not, just standard settings adjustments but no white balance or anything like that.

Also, I feel like I derailed this thread a bit my questions so I apologize. :scared:

SamuriHL

16th March 2021, 19:38

Not derailing the thread at all.

As for bt.1886, it's not as simple as "= 2.4".

https://kb.portrait.com/help/bt-1886-10-questions-10-answers

It'll approximate it, but it IS different.

VBB

16th March 2021, 19:42

With perfect black level, there is no difference mathematically between 2.4 and BT.1886.

Black level depends on the panel. Since you have a C7, like I do, you could try 51 and see if there's any glow. You need to do this in a completely dark room and with a 0% pattern. My C7 starts to glow at 53. With 52, there is a very faint dither pattern if you put your nose on the screen ;) So I settled on 51.

I would set both the TV and madVR to 2.4 gamma. Calibrated in this case simply means match whatever you set the panel to. Doesn't matter if it's professionally calibrated or eyeballed.

I for one welcome the derailing. No offense to the guys testing the curves ;) Much appreciated!