AVS Forum banner

3181 - 3200 of 9454 Posts

·
Registered
Joined
·
7,954 Posts
The scene in predator should be in your sample case. On full film is right where the helicopter shows up and the sky goes from light to dark. I came by to say that the 100 target peak nits, 65% desat, 500 nits desat looks really good to me. I am getting the detail back with the pop still there for color so far from testing I've seen in motion tests.
The sky goes from light to dark in the moment (or slightly after) the first time the white fonts appear on screen, right? The reason for that is the until the font appears for the first time, the whole scene is not very bright. So madVR tries to optimize for the limited brightness, to make the sky nicely bright. But then suddenly the font appears, and it's *MUCH* brighter than the rest of the sky. As a result, madVR has no choice than to adjust the tone mapping to account for the brighter font. An intelligent human being would say: Screw the font, continue to optimize for the sky. But madVR is a simple computer algo, so it can't possibly know if the font is very relevant or not. From madVR's view point, it *has* to optimize for the font because otherwise there'd be the risk of losing highlight detail. Consequently, the sky gets quite a bit darker, in that moment.

I don't think there's much I can do about it. Actually, we were discussing maybe adjusting to brighter measurements quicker than currently. If we do that, the sky might turn darker quicker than now. Which might actually look worse in this scene. So we'll have to keep this Predator scene in mind.

See timecode above, it's in Pacific Rim, not Pacific Rim Uprising, the same scene I've used many times to test highlights and I reported it to flash with an earlier build, that's why I didn't provide more details. You have to rewind 10-15 secs before that to give the time to the measured peak to reset, otherwise you don't get the flash.
I don't seem to be able to reproduce any "flash" here. Does the flash happen directly during the scene change? Is it the last frame of the previous scene which flashes, or the first frame of the following scene? And how would you describe the flash, if you step through the frames one by one. Is one frame much brighter than all the other frames? Or why does it look like a flash?

I am thinking the same, I prefer a 300-400 limit based on the 2 posted comparisons.
I'm more like 300-500, with 500 looking better to me than 300. But 400 might be an acceptable compromise. Which is about 200nits higher than the measured peak.

One for today, I hope to have more free time tomorrow:

@378nits, clipping, reference

I would say I prefer a 400-500 limit here (but very close to 500), the nits target does not seem to matter.

What do you think?
I'd say 500-600. But I agree that the nits target doesn't seem to matter that much. The measured peak seems to be more important.

On latest test build. I am satisfied. Thanks madshi and others for this great work.

target nits: 130 nits
tone mapping curve: BT.2390
color tweaks for fire and explosions: High
highlight recovery strength: low
measure each fram's peak luminescence checked.
output video in HDR format: unchecked.
desat control: 75% scaled to measured peak
desat control- measured peak limit: 600 nits
Thanks for your feedback. FWIW, compared to the other users, 75% scaled to measured peak is a bit higher than average. Average seems to be around 65% so far.

So I've been testing output to the JVC projector using the BT2020 colour space on the projector and madvr's basic settings that most people including Javs are using for the display settings/tone mapping settings and as a matter of interest I tested the playback of some normal BT709 blu-rays and these look surprisingly accurate to my untrained naked eye using the same settings. Can anyone tell me from their testing if normal blu-rays will show accurately using the BT2020 colour space on the projector with the BT2020 colour space and 2.4 power gamma curve selected in madvr due to the fact that the BT709 colour space effectively fits inside? My apologies if this is a silly question.
Not sure if it's an optimal configuration for BT.709 content, but it should work fine, I guess.

I am using Win 10, ver 1809. I have generated a debug log, please see the link below
Unfortunately the log is somewhat incomplete. It starts 2 minutes after playback has started. Probably the long got shortened because it grew too big. Can you please redo it, and close the media player already like 10 seconds after playback has started? Also please zip the log before uploading, thanks.

On my second issue where I do not get the Apply button if I want to test the new scale to peak options, I can confirm that all other controls work fine and I have only once instance of madVR with no other madVR files scattered around on my PC. Additionally, even if I reset to defaults, losing all profiles, I cannot have the scale to peak option to stick. It always reverts back to scale to 10000 nits!
My best guess is that when installing the latest test build, you might have accidently failed to replace the mvrSettings32/64.dll files, maybe because madVR was still running when you tried to overwrite the files? Please download the latest test build again and overwrite the files once more, with the "mad" tray icon application (madHcCtrl.exe) closed and no media player running.

I didn't had too much time for testing but far on latest test build my preferences are

target nits: 250 nits
desat control: 60% scaled to measured peak
desat control- measured peak limit: 500 nits
Those settings works for me so far.

I like have a little bit more saturation but desat control: 65% scaled to measured peak also works and it gives you a little more highlights.
For the highlights/saturation balance my choice is often desat control- measured peak limit: target nits +300 mostly. If desat control- measured peak limit higher more saturation is added. Looks like orange or orange-like colors affected most but i see on red,yellow and green. For my eyes blue change is minimal. Skin tones also affected.

Also less saturation have good efect on some scenes such as sun light on the rocks like samsung wonderland demo.
If highlights on the rocks more saturated you get less hdr efect because two color diffrence is less. So desat control: 65% scaled to measured peak is better here.
Thanks. That's near to what the majority of users have voted for, so atm I think it will be 65%. Not completely sure on the limit yet, though.

I haven't had too much time to test this week but I concur with the 65% desat and 500 limit. I tested that out over the weekend with Ant-man and the Wasp and it looks damn good. Did not get to take screenshots.
Thanks.

Here's another good example (I think)

From Murder on the Orient Express (look at the saturation level of the yellow lamp shade)

Reference 222 clipping

My preference is around 500.
That is strange, my highest limit would be 400 here, with probably a balanced result around 350.

The ideal limit for me seems to be the measured peak + x, but I am not 100% sure.
My lowest limit would be 400 here, but I might prefer 500.

I agree that we will probably end up using "limit = measured peak + x". I'm not sure yet what "x" will be. I feel that using a higher number would be "safer". After all, the original "scale to measured peak" was always using a limit of 600nits, and it looked fine. I'm thinking "x" may end up something like 222nits... :p

I would say try to check out 2fast 2furious if anyone of you can. Settings at 100 target peak nits, 65% desat, 500 desat some of the films looks like an older kind of film in some shots. Looks good in others, but some look like an old film reel in some shots. Also, I'm not sure if this setting is losing contrast or not.
I'm not completely sure how to understand your comment. Is looking like an older kind of film good or bad? If it's bad, then which other settings combination would you suggest instead?

Sorry if this question is off topic, wasn't sure where to ask it, but its related to MADVR. I have always ran my PC and all my consoles at Full Range RGB and calibrated the projector to expect full range RGB. But with HDR games needing limited range 10bit to achieve @ 60hz I am guessing I need to now calibrate to limited range to accommodate that.

Uses\Plan:
Movies using MADVR will run at 24HZ 4:4:4 10bit Y’CbCr
PC desktop will run at 60HZ 4:4:4 8bit Y’CbCr
PC NON HDRgames will run at 60hz 4:4:4 8bit Y’CbCr
Console game will run at 60hz 4:2:0(2) 10bit Y’CbCr
PC HDR games will run at 60hz 4:2:0(2) 10bit Y’CbCr (force through fullscreen game mode)

So wouldn't I need to calibrate the projector at Y’CbCr range (limited) and have to use Y’CbCr 4:4:4 output for the GPU on the PC? For MADVR would I output RGB Full and let the GPU do the conversion at this point?
I neither like nor recommend YCbCr output. I always strongly recommend RGB. But in the end, every TV is different. Some TVs may not handle RGB signals as well as they should. So maybe there's a good reason to use YCbCr output. In any case, unless you send me your TV for testing, I don't know what to recommend to you. You're in the best position to test this yourself and pick the settings combination which looks best for you.

In any case, this doesn't seem to be a topic related to improving madVR's tone mapping algorithm? Which is what this thread is all about. So I'd kindly ask you to follow-up on this on another madVR related thread here on AVSForum or doom9, so we can keep this thread dedicated to improving madVR's tone mapping algorithm. Thanks!
 

·
Premium Member
Joined
·
9,837 Posts
I don't seem to be able to reproduce any "flash" here. Does the flash happen directly during the scene change? Is it the last frame of the previous scene which flashes, or the first frame of the following scene? And how would you describe the flash, if you step through the frames one by one. Is one frame much brighter than all the other frames? Or why does it look like a flash?
It's the same issue I described a few builds ago when you introduced desat settings and we had issues at 1000nits target peak.

If you start playback at the beginning of the chapter (the timecode above), there is no flash, because the measured average is locked to the brighter picture from the beginning.

If you start playback 10-15 secs before the beginning of the scene, during the darker scene where they talk, the rolling average is lower when the brighter scene starts, and I assume that this is what triggered the issue: a few seconds into the brighter scene, just before or as people are walking by in the foreground, there was a brief and sudden brightness peak, probably more than a single frame.

I tried to reproduce it with 500 target peak, 65% desat and limit of 300nits (the settings that produced it during my last testing) and I couldn't.

In any case, I think this will be resolved if you use a limit of measured peak + x, because I could get rid of it by selecting a limit of 350nits. I selected 400nits afterwards to be safe, and never saw it again. I am now set to 480nits target peak for that profile and I haven't seen it with my current settings of 65% desat and 500nits limit.
 

·
Registered
Joined
·
1,209 Posts
My lowest limit would be 400 here, but I might prefer 500.

I agree that we will probably end up using "limit = measured peak + x". I'm not sure yet what "x" will be. I feel that using a higher number would be "safer". After all, the original "scale to measured peak" was always using a limit of 600nits, and it looked fine. I'm thinking "x" may end up something like 222nits... :p
What about this one?

@200nits, clipping, reference




@100nits, 65% desat - limits from 100 to 1000 (various steps)



Between 300 and 400 for me, 400 max. The biggest change is between 200 and 300.

One more:

@506nits, clipping, reference




@100nits, 65% desat - limits from 100 to 1000 (various steps)



Thoughts?

I am thinking 600 is just not enough and 800 is too much.
 

·
Registered
Joined
·
410 Posts
Here's one from Interstellar, which has warm whites. (IIRC, I read somewhere that the film negative that Christopher Nolan uses, has a tendency to produce warm whites)



Here's the clipped reference at 382 nits.







Here are the images with a target peak nits of 300, so there isn't a great deal of tone mapping occurring, and the temperature of the sky should remain fairly close to the reference image.



300 vs 400 vs 500 vs 600









In this instance, the difference between 400 to 500 is more noticeable than 300 vs 400 imo.


And given that a drop in 82 nits from 382 to 300 isn't that great, I think the 300 limit changes the temperature of the white too much compared to the reference. I think 500 is more representative of the correct white temperature imo




edit: just seen neo's comments post re: 506 nits. TBH, I haven't looked at any images that high when testing the limit. So I'll look for content in that range.
2nd edit: upon looking at those shots from Oblivion, it might be too had to tell given the large discrepancy in the brightness level from 500 nits to 100 nits.
 

·
Registered
Joined
·
378 Posts
ok, with what neo is talking about, I think, most of the film with 2fast 2furious looks good at 500 measured peaks, but some scenes look more correct at 600 to 800 measured peak nits, like it's too desaturated at 500 measured peak nits.

I think I am going to stick with 600 measured peak nits for right now with 65%desat at 100 target peak nits because it just looks better to me.

The sky in 2fast 2furious looks a lot better at 500 measured and a little too saturated at 600 to 800 measured peak nits, but the other thing I was talking about, desaturation, looks better at 600 to 800 measured peak nits.
 

·
Registered
Joined
·
1,209 Posts
Here's the clipped reference at 382 nits.
And given that a drop in 82 nits from 382 to 300 isn't that great, I think the 300 limit changes the temperature of the white too much compared to the reference. I think 500 is more representative of the correct white temperature imo
I prefer 500 here too, 400 is clearly not enough and 600 is a little too much IMO.

In this instance, the difference between 400 to 500 is more noticeable than 300 vs 400 imo.
I noticed that too for the other comparisons, the biggest difference seems to be at measured peak + 100.

It doesn't say anything about how much we should go over the measured peak, but maybe that the formula should be based on the measured peak.


Last one?

@291nits, clipping, reference




@100nits, 65% desat - limits from 100 to 1000 (various steps)



This one is very interesting I think. Don't look only at the left side of his face, but also at the desaturation of the green color in the background (right side of the image).

Preferred: between 400 and 500. Not under 400 and big max 500.

Measured peak + 100-200 would be ok for me (but no more, then it looks overcooked IMO).


Edit: Same opinion with a 200nits target:

@200nits, 65% desat - limits from 100 to 1000 (various steps)

 

·
Registered
Joined
·
13 Posts
Unfortunately the log is somewhat incomplete. It starts 2 minutes after playback has started. Probably the long got shortened because it grew too big. Can you please redo it, and close the media player already like 10 seconds after playback has started? Also please zip the log before uploading, thanks.


My best guess is that when installing the latest test build, you might have accidently failed to replace the mvrSettings32/64.dll files, maybe because madVR was still running when you tried to overwrite the files? Please download the latest test build again and overwrite the files once more, with the "mad" tray icon application (madHcCtrl.exe) closed and no media player running.
I have redone the log and zipped it, hopefully it will provide you with the necessary data this time. The link is:
https://wetransfer.com/downloads/6235bbd7dda008f056376f9a1454b1dc20181017181146/1c3de6fafecf32e4cccef520a111818320181017181146/4a168a
On the second issue, you are of course right. Windows 10 1809 works in mysterious ways I can conclude, because before a few minutes ago I copied this build in the past 3 days like 20 times directly from the archive without Windows complaining or showing anything suspicious, files were showed copied over. This time I unzipped it and copied over and can report that the new scale to peak features are working. Finally. Thanks for the suggestion.

Edit: the first issue was in fact related to the second. I have the histograms working now. I discovered as I started testing... Please do not waste anymore time on this. Thank you for your support showed!
 

·
Registered
Joined
·
1,209 Posts
Last one:

@629nits, clipping, reference




@200nits, 65% desat - limits from 100 to 1000 (various steps)




This seems confirm what I thought:

- the limit should not be under the measured peak, because 600 here is visibly not enough
- the target nits does not matter (I tested from 100 to 300, same conclusion)
- the biggest difference (the difference that matters) is when the limit is above the measured peak, here it is between 600 and 800
- limit = measured peak + x works

For me, 800 is the max tolerable limit here (~200 above the measured peak). 1000 is clearly way too much IMO.
Preferred: probably around 700, but this limit is not available to compare unfortunately.

Peace y'all.
 

·
Registered
Joined
·
7,954 Posts
the first issue was in fact related to the second. I have the histograms working now.
Good to hear!

In any case, I think this will be resolved if you use a limit of measured peak + x, because I could get rid of it by selecting a limit of 350nits. I selected 400nits afterwards to be safe, and never saw it again.
Ok, then I'll just ignore that issue for now.

What about this one?

@200nits, clipping, reference

Between 300 and 400 for me, 400 max. The biggest change is between 200 and 300.
Between 400 and 500 for me. It seems I'm always about 100nits higher than you, for some reason.

One more:

@506nits, clipping, reference
I can hardly see a difference here. Not enough to make a judgement.

Here's one from Interstellar, which has warm whites. (IIRC, I read somewhere that the film negative that Christopher Nolan uses, has a tendency to produce warm whites)

Here's the clipped reference at 382 nits.

Here are the images with a target peak nits of 300, so there isn't a great deal of tone mapping occurring, and the temperature of the sky should remain fairly close to the reference image.

300 vs 400 vs 500 vs 600

In this instance, the difference between 400 to 500 is more noticeable than 300 vs 400 imo.

And given that a drop in 82 nits from 382 to 300 isn't that great, I think the 300 limit changes the temperature of the white too much compared to the reference. I think 500 is more representative of the correct white temperature imo
Actually, I prefer 600 here. As Neo-XP mentioned, the biggest difference seems to be when reaching measured nits + 100nits. But I think I just like it a touch more saturated, so I'm at about +200-300nits. I could live with +200nits or +300nits, but I think my personal favorite would be +250nits.

Last one?

@291nits, clipping, reference

@100nits, 65% desat - limits from 100 to 1000 (various steps)

This one is very interesting I think. Don't look only at the left side of his face, but also at the desaturation of the green color in the background (right side of the image).

Preferred: between 400 and 500. Not under 400 and big max 500.

Measured peak + 100-200 would be ok for me (but no more, then it looks overcooked IMO).
As you say, the biggest difference happens at +100nits. After that, higher values only add a little bit of saturation, each. With this specific image I'd be fine with +200nits, but I could just as well live with +300nits here, too. Or even more.

-------

One thing I'm wondering about: If we use "limit = measured nits + x", does that end, anywhere? Obviously if the measured peak is 10,000nits, "x" will be 0. But e.g. if the measured nits is 4000nits, then should be still add "x" on top? Is there a certain "measured nits" threshold, above which we should not add an "x"? Or should we always add it?
 

·
Registered
Joined
·
1,209 Posts
One thing I'm wondering about: If we use "limit = measured nits + x", does that end, anywhere? Obviously if the measured peak is 10,000nits, "x" will be 0. But e.g. if the measured nits is 4000nits, then should be still add "x" on top? Is there a certain "measured nits" threshold, above which we should not add an "x"? Or should we always add it?
Well.. it works (for me at least) for a measured nits peak up to 629nits (see my post above yours). What do you think about this comparison? What is your your preferred limit?

If you want us to test for higher measured nits peaks, we will need a new build with higher limits I guess :D

Or better, a build with limit = measured peak + 50, 100, 150, 200, 250, 300... so we can find the "sweet spot" more easily.
 

·
Registered
Joined
·
7,954 Posts
Well, true, the test build tops out at 1000nits limit. However, you could still test with e.g. a 2000nits measured peak scene and double check that 1000nits is really better for that scene compared to 800nits. If it is, then at least we know that the limit should be allowed to reach at least up to 1000nits, maybe more.

For many of these comparisons, I find it really hard to pick the best option. Clearly anything below or directly at the measured peak looks too desaturated to my eyes so far. But above that I'm often not sure about the best value.

Anyway, if you guys can confirm that the "+ x" algorithm holds true until 1000nits, then I can easily create another test build with measured peak + 50, 100, 150 etc. But it would be nice to double check that first. Maybe the limit should never be higher than 800nits? I kinda doubt it, but who knows?
 

·
Registered
Joined
·
1,209 Posts
Well, true, the test build tops out at 1000nits limit. However, you could still test with e.g. a 2000nits measured peak scene and double check that 1000nits is really better for that scene compared to 800nits. If it is, then at least we know that the limit should be allowed to reach at least up to 1000nits, maybe more.
That will be very difficult to find I guess, because if there is a difference, it will not be much visible if we can't test with a limit above the measured nits peak.

This is the best I can do with the current build:

@787nits, clipping, reference




@100nits, 65% desat - limits from 100 to 1000 (various steps)



As you can see (I hope), it still works for a measured nits peak of ~800. The biggest difference is between limits of 800 and 1000.

Preferred limit: 1000.

I can not test with higher measured nits peaks, because I would have to check if a limit of 1200 for instance still works for a measured peak of 1000nits.

I am done testing for this build on my side I guess, I hit the limit :)
 

·
Registered
Joined
·
5,364 Posts
One thing I'm wondering about: If we use "limit = measured nits + x", does that end, anywhere? Obviously if the measured peak is 10,000nits, "x" will be 0. But e.g. if the measured nits is 4000nits, then should be still add "x" on top? Is there a certain "measured nits" threshold, above which we should not add an "x"? Or should we always add it?
That's a very good question. We need a way to test this in the real world and take a look at what makes sense, no? I do agree with your ~+250nits idea, though. But here's another question....at the higher end as you approach whatever we decide the "cutoff" will be to not apply "x" on top, should "x" be smaller? Like a curve?

EDIT: Missed that you posted a new build....so at least we know how to test but my question about a curve as we approach x=0 stands.
 

·
Registered
Joined
·
1,209 Posts
Alright, here's a modified settings dialog, which allows setting the limit up to 4400nits:

http://madshi.net/madHcCtrlLimit.zip
Thanks, I hope to have some free time tomorrow to find good scenes to test with higher measured nits peaks.

A quick one for now:

@1064nits, clipping, reference



@100nits, 65% desat - limits from 200 to 2200 (+400 steps)



No visible change with limits under the measured peak, then visible change between 1000 and 1400, 1400 and 1800, 1800 and 2200... (linear change it seems).
 

·
Registered
Joined
·
1,209 Posts
@1244nits, clipping, reference




@200nits, 65% desat - limits from 100 to 4400 (various steps)




@100nits, 65% desat - limits from 100 to 4400 (various steps)




No visible difference between 100 and 1200, but then a little more saturation with each +400 step.

1600 (~400 above the measured nits peak) seems already too high IMO (losing too much luminance).
1400 (~150 above the measured nits peak) might be not enough.


Just my opinion, so what do you think ?
 

·
Registered
Joined
·
1,479 Posts
Discussion Starter #3,197
@1244nits, clipping, reference


No visible difference between 100 and 1200, but then a little more saturation with each step.

1600 (~400 above the measured nits peak) seems already too high IMO (losing too much luminance).
1400 (~150 above the measured nits peak) might be not enough.
Do you still think that scaling to the peak nits + x value for the limiter, is significantly better than scaling to 100% 10000nits?
Always?
 

·
Registered
Joined
·
1,479 Posts
Discussion Starter #3,198 (Edited)
Also if 65% desat is good for a limit equal to peak nits + x.

Maybe the x will akways change in % of the peak. And at 10000nits peak, since you can't say peak + x, you would have to pick a desaturation lower than 65%.

Probably you can achieve the same results with a lower desaturation than 65% but this time directly to the peak (without + x).
@Neo-XP
Could you look if a lower desaturation than 65% + x, can give you the same result but directly scaled to peak nits?

Maybe 50%? Or 40% to peak?

Thank you!!
 

·
Registered
Joined
·
943 Posts
Guys quick question.. For those who have a JVC projector when tone mapping with madVR do you keep the JVC gamma at custom and then 2.2 as well as I the madVR calibration settings? The reason I ask is yesterday I was experimenting with different gamma settings on the JVC and found the setting B to have much more pop and contrast which gave it more of an HDR look than the 2.2 setting in the JVC gamma setting.. Would appreciate if someone with a JVC could comment.. Thank you..

Sent from my moto g(6) using Tapatalk
 

·
Registered
Joined
·
1,479 Posts
Discussion Starter #3,200
Guys quick question.. For those who have a JVC projector when tone mapping with madVR do you keep the JVC gamma at custom and then 2.2 as well as I the madVR calibration settings? The reason I ask is yesterday I was experimenting with different gamma settings on the JVC and found the setting B to have much more pop and contrast which gave it more of an HDR look than the 2.2 setting in the JVC gamma setting.. Would appreciate if someone with a JVC could comment.. Thank you..

Sent from my moto g(6) using Tapatalk

1) did you measure your jvc to make sure than 2.2 is really 2.2 and not 2.0 or lower because of the famous jvc gamma drop?

2) generally (also for sdr content) going from 2.2 to a gamma 2.4 gives you extra picture depth.
 
3181 - 3200 of 9454 Posts
Top