AVS Forum banner

2741 - 2760 of 9466 Posts

·
Registered
Joined
·
5,364 Posts
Yes I used measured 2 when testing the tone mapping.

Sent from my Pixel XL using Tapatalk
 

·
Registered
Joined
·
379 Posts
the scene I'm talking about in Predator is a fade into the helicopter at the very beginning of the movie. I haven't seen any other weird problems in anything else. Predator always seems to be my go-to for checking color, contrast, balance and so on.
 

·
Registered
Joined
·
11 Posts
What target nits are you using?
What peak nits is the scene at?

Does the rolling avg peak nits change between the normal playback pause and the reactivation of measurement on your last picture?
Target nits for madVR: 125

when paused during playback: measured 87nits, avg 182 nits
after disabling and re-enabling measure while still paused: measured 87nits, avg 87 nits
 

·
Registered
Joined
·
7,954 Posts
I have a similar find as Neo-XP. I am using Measure2 (Measure1 didn't work for x64 as you pointed out).

This is from Last Jedi 00:48:42.586

This wasn't at a scene change boundary. Any ideas?
Actually, although this is not a scene change in the conventional (movie script) point of view, the movie does cut back and forth between Luke's face (or Luke and Rey) with a bright sky in the background, and then Rey, sitting on the rock, with a much darker background. Whenever the bright sky is in the scene, peak brightness measures about 270 Nits. Then when the scene cuts to Rey on the rock, peak brightness jumps down to about 90 Nits. So there's a 3x Nits difference between the different camera perspectives. The rolling average needs about 3 seconds to fully adjust to a new camera perspective.

From madVR's point of view, these are all (micro) scene changes. So once I implement scene change detection, these problems should (hopefully) go away.

Of course another thing we can test is how quickly the rolling average should react. Currently it's 3 seconds. I could make it less (or more). I may offer a test option for that, but only once I implemented scene change detection.
 

·
Registered
Joined
·
7,954 Posts
Actually, I've changed my mind. Here's a new test build:

http://madshi.net/madVRhdrMeasure3.zip

Still no scene detection. Here are the changes:

1) I'm now rounding up measurements to 100nits, which seems to help avoiding some artifacts.
2) There's a new option which lets you choose the rolling average reaction time.

Please let me know which reaction time works best for you. If you make it too fast, you'll get flickering with some content. The scene from Murder on the Orient Express is a *great* scene to test for flickering. If you turn the rolling average off, this scene flickers like crazy. FWIW, all previous builds used a hard coded reaction time of 3 seconds.
 

·
Registered
Joined
·
379 Posts
fade in weirdness on predator is gone. Thank you. Not sure how to set this btw so I will wait to see what everyone else does. Also, I'm curious as to what everyone is setting their saturation to.
 

·
Registered
Joined
·
1,479 Posts
Discussion Starter #2,747 (Edited)
Actually, I've changed my mind. Here's a new test build:

http://madshi.net/madVRhdrMeasure3.zip

Still no scene detection. Here are the changes:

1) I'm now rounding up measurements to 100nits, which seems to help avoiding some artifacts.
2) There's a new option which lets you choose the rolling average reaction time.

Please let me know which reaction time works best for you. If you make it too fast, you'll get flickering with some content. The scene from Murder on the Orient Express is a *great* scene to test for flickering. If you turn the rolling average off, this scene flickers like crazy. FWIW, all previous builds used a hard coded reaction time of 3 seconds.
Thanks! :) I am going to play right away :)

About the rounding up to 100 nits, what does it effctively means with 300nits target nits?
Before with rounding up to 600nits, we were over-compressing the highlights for peak up to 600nits.

So what does it mean with 100nits lower than your target nits?

Do you use it in your INTERN rollling peak avg, so that any frame peak value below 100nits are overwritten with 100?
If not, this would be a simple countermeasure to the fade to black until we get true scene detection.
But in this case, we could use this strategy for anything lower than the target nits. So the measure peak would clip at the lower limit of your target nits.

Of course, when you re-implement the clipping below target nits with a smoother transition with tone mapping, we may to revisit that?

But in any case scene change will be needed when we go full dynamic and adjust the target nits on the fly or something like that.

EDIT:
I love this enhanced measure peak . Difference with no measure is night and day now! :eek:
I also confirm that the avg peak never gets below 100nits, even if the current peak stays below 100nits for a longer time. So I did understand your fix correctly and yes it will definetely help against black out scenes.

I tested a bit with 1s rolling avg. I saw nothing bad on the screen. Needs more testing.
 

·
Registered
Joined
·
1,209 Posts
1) I'm now rounding up measurements to 100nits, which seems to help avoiding some artifacts.
The "weird" artifacts like the one in La La Land scene 1 are gone with this now.

2) There's a new option which lets you choose the rolling average reaction time.

Please let me know which reaction time works best for you. If you make it too fast, you'll get flickering with some content. The scene from Murder on the Orient Express is a *great* scene to test for flickering. If you turn the rolling average off, this scene flickers like crazy. FWIW, all previous builds used a hard coded reaction time of 3 seconds.
I do not notice any bad effect at 1s. The best would be the lowest delay until we notice flickering I guess (because it can be bad too if the algo takes too much time to adapt).

One flickering example:

Flickering in La La Land: http://www.mediafire.com/file/wskhd4x5k0jzkw0/La_La_Land_scene_2.mkv/file

I still notice the flickering at 0.75s, not at 1s, but I do not know if I am very sensitive to this or not.

To test a scene change (for future scene detection algo):

Looking outside from 10 Cloverfield Lane: http://www.mediafire.com/file/xu7auinajlxbgx3/10_Cloverfield_Lane_scene_1.mkv/file

Lights in Red Sparrow: http://www.mediafire.com/file/lil61rnbj2v4i4v/Red_Sparrow_scene_1.mkv/file

Thanks madshi.

Edit: it seems that the modifications do not apply until I untick + apply and tick + apply the "measure each frame's peak luminance" box.
 

·
Registered
Joined
·
787 Posts
Great job @madshi and others helping :D

I'm sorry for being a knob, but what should be ideal target peak nits set to on my JVC projector (RS420) when doing HDR to SDR?
my projector screen has about 50 nits. I use 200 nits as my target to give picture a little pop. Is 200 nits too low or too high? @Javs what are you using at this time

Also using last night build I cannot see any difference in saturation even when going from one extreme to another -50% or +50% at least to my naked untrained eye.
 

·
Registered
Joined
·
379 Posts
Great job @madshi and others helping :D

I'm sorry for being a knob, but what should be ideal target peak nits set to on my JVC projector (RS420) when doing HDR to SDR?
my projector screen has about 50 nits. I use 200 nits as my target to give picture a little pop. Is 200 nits too low or too high? @Javs what are you using at this time

Also using last night build I cannot see any difference in saturation even when going from one extreme to another -50% or +50% at least to my naked untrained eye.
I can't see the difference either, that's why I was asking others.
 

·
Registered
Joined
·
8,456 Posts
Great job @madshi and others helping :D

I'm sorry for being a knob, but what should be ideal target peak nits set to on my JVC projector (RS420) when doing HDR to SDR?
my projector screen has about 50 nits. I use 200 nits as my target to give picture a little pop. Is 200 nits too low or too high? @Javs what are you using at this time

Also using last night build I cannot see any difference in saturation even when going from one extreme to another -50% or +50% at least to my naked untrained eye.
I am using 330 nits target at probably about ~60 nits output. I decided to stop down my iris a lot to -8 in low lamp, and honestly I really like it. More punch than SDR and very nice contrast. I watch SDR at -12.
 

·
Registered
Joined
·
787 Posts
I can't see the difference either, that's why I was asking others.
I'm glad I'm not the only one :)

I am using 330 nits target at probably about ~60 nits output. I decided to stop down my iris a lot to -8 in low lamp, and honestly I really like it. More punch than SDR and very nice contrast. I watch SDR at -12.
thanks for your prompt feedback. ;) I will try 250 nits to be closer to you. :D At the moment IRIS is set to -7 (low lamp). For SDR I have it at -10 (low lamp). For gamma I had it at 2.4 (both JVC and madVR), but recently I've changed it to 2.2 (both JVC and madVR). I feel it gives me more shadow details.

On the side note, my NVidia 399.xx drivers ((gtx1060(6gb)) + win10 1803 (latest updates)) sometimes loses 12bit RGB settings (and defaults them to 8bit RGB). To fix it I have to go to 23Hz and change it to 12bit. It doesn't happen on every system start, so I don't know what is causing it. Like @madshi said we should be all using windows 8.
 

·
Registered
Joined
·
8,456 Posts
Like @madshi said we should be all using windows 8.
Sorry but my computer is not built only for MadVR...

I happen to like Windows 10 and dont have any problems with it.
 

·
Registered
Joined
·
124 Posts
I am using 330 nits target at probably about ~60 nits output. I decided to stop down my iris a lot to -8 in low lamp, and honestly I really like it. More punch than SDR and very nice contrast. I watch SDR at -12.
I basically just steal all of your settings and wound up being able to watch HDR content, in low lamp mode, that looks great. A little black crush, and sometimes when i flip back to regular HDR high lamp mode there is more pop in the colors but I think I have accomplished the goal...With the 790 we will always be limited by lumens, I guess, so I have a hard time imagining it is going to get any better than what I have for a while.

I will be continuing to rip off all your guys' settings whenever you figure stuff out. Cheers!
 

·
Registered
Joined
·
7,954 Posts
About the rounding up to 100 nits, what does it effctively means with 300nits target nits?
Before with rounding up to 600nits, we were over-compressing the highlights for peak up to 600nits.

So what does it mean with 100nits lower than your target nits?

Do you use it in your INTERN rollling peak avg, so that any frame peak value below 100nits are overwritten with 100?
Yes, exactly.

But in this case, we could use this strategy for anything lower than the target nits. So the measure peak would clip at the lower limit of your target nits.
That would be possible, but if we don't "allow" any measurements lower than the target nits then it will be very hard for the average to ever go down to the exact target nits, which would make it impossible for madVR to ever get into "clipping" mode, even if the true measurements are much lower than the target nits. So I think it's better to allow measurements to be a bit lower than the target nits.

I love this enhanced measure peak . Difference with no measure is night and day now! :eek:
I also confirm that the avg peak never gets below 100nits, even if the current peak stays below 100nits for a longer time. So I did understand your fix correctly and yes it will definetely help against black out scenes.

I tested a bit with 1s rolling avg. I saw nothing bad on the screen. Needs more testing.
Glad to hear you like it!

I do not notice any bad effect at 1s. The best would be the lowest delay until we notice flickering I guess (because it can be bad too if the algo takes too much time to adapt).
I agree, in theory faster reaction times should be better. I definitely can still see some flickering with 0.75 seconds. At 1.00 seconds it might be gone, but I'm not completely sure yet. I may still see a slight hint of flickering right at the start of your Orient Express sample (more so than in your La La Land flickering sample). I'm a bit worried about going too low with this setting. Maybe we should rather play it save. Flickering can be really annoying, even if the amount of flickering is so low that it's almost unconscious.

Thanks for the samples!

Edit: it seems that the modifications do not apply until I untick + apply and tick + apply the "measure each frame's peak luminance" box.
This might be another case of me forgetting to recompile 64bit. The settings apply immediately in 32bit, maybe the immediate apply is broken in 64bit.
 

·
Registered
Joined
·
91 Posts
I just recently went back to Windows 8.1, and not planing to reinstall Windows 10 (probably never) as it is a total mess with nvidia drivers at the moment. If you have a dedicated HTPC, the best option is 8.1 OS. Love 8.1 so far.
On the side note, my NVidia 399.xx drivers ((gtx1060(6gb)) + win10 1803 (latest updates)) sometimes loses 12bit RGB settings (and defaults them to 8bit RGB). To fix it I have to go to 23Hz and change it to 12bit. It doesn't happen on every system start, so I don't know what is causing it. Like @madshi said we should be all using windows 8.
 

·
Registered
Joined
·
18 Posts
Fullscreen Exclusive (OSD wise anyway) seems to be working again with Windows 10 1809, they must have fixed what they broke.

Shows 60 Hz in the composition rate, although its 23.976, had that before if I remember.

Unsure about much else madVR wise as I am limited to 8Bit SDR with my setup, will maybe have a play with later Nvidia drivers and see if custom resolutions are working. (more than likely the Driver as we know).

I only did the insider preview again as I think it was a case of self harm and if it broke I was going to 8.1 myself, but it looks like there are other improvements, as Explorer was annoying me with refresh rates of files, ie. multiple FLAC albums, but its really snappy again.

I'm sure something will bite me soon. :)
 

·
Registered
Joined
·
1,479 Posts
Discussion Starter #2,758 (Edited)
Yes, exactly.


That would be possible, but if we don't "allow" any measurements lower than the target nits then it will be very hard for the average to ever go down to the exact target nits, which would make it impossible for madVR to ever get into "clipping" mode, even if the true measurements are much lower than the target nits. So I think it's better to allow measurements to be a bit lower than the target nits.


Glad to hear you like it!
Yes I agree that we actually want to go lower to activate clipping mode.
Plus with the coming dynamic HDR, we definetly want to be able to ge lower than the user given display target nits.

About that, I have a simple idea we would try almost right away.

Example with user based display target nits 300:

Right now we have the the measured single frame target nits measurement, and the short rolling avg (1s to 3s) to adjust the high end and decide to compress just as much as needed accordingly.

What we could do is introduce a second rolling avg with a long avg time like 60s which would be synchronized with the display target nits.
If a movie is dark all the time (Bourne, Arrival, Han Solo, etc...), this will quickly adjust itself automatically to something pretty stable like 150nits display target nits.

For the frame which have higher peak nits, the first rolling avg will continue to provide the information how much we need to compress over 150nits.
And if the short rolling avg says that the peak nits is lower than 110nits for example, then it will clip at 150nits.

Of course we can do the same you did for the short rolling avg, and provide the lowest value allowable for the 2nd big rolling avg.
Like for a user target nits of 300nits, we could adjust to a minimum of half the value of 300=150nits.
We could also adjust the target nits with the 2ndf rolling avg to be even higher than the user input, but it's not the prioriy at the moment.

Priority is to help with dark movie or dark scenes.

:)

EDIT: when we first start a movie, we need to come up with a safe value of the 2nd big rolling avg. We could take "user input display target nits=300" as the "others" measured peak value until the large time (60s) is finally fully filled with real measured value.
Like after 5s of movie at 100 peak nits mesured, the big rolling avg would be= (5*100+(60-5)*300)/60=283 nits.

Like this it shoud go smoothly and slowly to a lower target nits.
Of course 60s is only an example, maybe we need 5min instead. :)
 

·
Registered
Joined
·
410 Posts
Just providing some feedback on the desaturation options:


On the whole, I was leaning from -10/-15 to default (there were greater deviations in some scenes). However, there were a few scenes i.e. Pan sunset/ the green spear in BvS, where -15 was perhaps too much. So, I wouldn’t go lower than -10, which might be my preference, but I’d be perfectly content with the current default
 

·
Registered
Joined
·
787 Posts
Sorry but my computer is not built only for MadVR...

I happen to like Windows 10 and dont have any problems with it.
I hope the latest October release 1809 will behave :)

I'm very happy with SDR and HDR madVR output. However I'm lacking color saturation. The picture is a tad too dull for my likings. What are your colour settings? I wish I could boos it a bit. I don't see much difference between JVC at +10 vs +40.
 
2741 - 2760 of 9466 Posts
Top