madVR - high quality video renderer (GPU assisted)

[Shiandow]

Originally Posted by bacondither

I extracted 16-bit pngs from Ver Greeneyes "gradient.mkv" he posted some while back. And then i took some screenshots from madvr ditherd down to 8-bit. It´s frame 89, and it is really dark. All imgs have their histogram capped at 10. So we are looking at the ~7 lowest steps of a 8-bit image(256 steps).

Source 16-bit
No Dither

ED Gamma light
ED Linear light

Random dither

The 16-bit source is a tiny bit brighter becouse i had to process it in another program(krita) that had support for >16-bit images and prosessing.
Kritas level controls did behave a bit strange.

Dithering in linear light gives a more accurate image to the source.

You need to be very careful with comparing the accuracy on brightened images. If you brighten it the wrong way then these kind of comparisons can easily become useless.

I think that in this case the correct thing to do would be to convert the images to linear light, then multiply the values by some constant, and then convert back to nonlinear light. Even then I am not 100% certain that this actually gives you a fair comparison since linear light may not necessarily seem linear to humans. For the most objective comparison you should probably blur the images in linear light, in that case a 'perfectly' dithered images should be identical to the original.

I have no way of verifying if this is the way it was done and I don't know if there even exists any software which performs brightening in this way.

[nevcairiel]

Originally Posted by trip_let

Sorry to interrupt the current discussion, but does anybody know what order madVR processing is done in relation to pixel shaders?

Specifically, are post-resize pixel shaders before or after madVR dither / ED?

Dithering is the very last step, after everything else, including after custom pixel shaders.

[nevcairiel]

Originally Posted by 6233638

The "gamma light" image looks a lot closer to the "source 16-bit" image to me.
Linear light dither also has obvious banding, similar in width to each band in the "no dither" image.

Are you sure you didn't mix them up?
Gamma Light seems to have the banding, and linear also looks closer to the original to me, although the end of the gradient doesn't match perfectly on either.

[6233638]

Originally Posted by nevcairiel

Are you sure you didn't mix them up?
Gamma Light seems to have the banding, and linear also looks closer to the original to me, although the end of the gradient doesn't match perfectly on either.

Comparing Bacondither's images, gamma definitely seems closest to the 16-bit source.
Those three images were used as the source for this comparison:



Black point is closest, and the gamma image values track closest to the 16-bit ones throughout the range. (neither are close in the upper range, but below about 50% gamma seems better)

[madshi]

Interesting. So should I trash the linear light build and stick to the gamma NL6 build? Opinions welcome.

[har3inger]

Sorry for the tangent, but I have a quick update on my troubleshooting with Directcompute dithering.

Given that my laptop uses two GPUs, I decided to try what would happen if I disabled the intel 4000 HD.

Well, turns out madvr will try to run in software mode or something, as simple playback of 8 bit content without scaling slows down to a <1 fps slideshow. However, direct compute dithering now "works", in that it doesn't result in a black screen. I am still unsure if it's automatically ignoring directcompute settings in absence of GPU acceleration, or if I've somehow bypassed a bug in the drivers. I don't know why mpchc/madvr won't use my discreet gpu in the absence of the integrated one, but I did notice that madvr is displaying through an "unknown generic monitor" with an orange icon rather than the generic monitor that indicates my laptop screen.

This leads me to believe that there is likely something wrong in the 13.12 catalyst drivers or the OEM intel 4000 drivers I'm using. Updating the 4000's drivers is tricky because every installer I have found refuses to install. I doubt the 13.12 drivers are the culprit, since virtually no one else has my problem.

[GREG1292]

Just got done comparing 7,LL,NL6 and from my projector
Linear Light is the clear winner. Still has the pop and
a veil has been lifted. Nice work

[Shiandow]

Originally Posted by madshi

Interesting. So should I trash the linear light build and stick to the gamma NL6 build? Opinions welcome.

In my opinion it is most likely that something, somewhere has gone wrong. Unless I made a serious mistake in my reasoning somewhere the linear light build should be brighter for values near black.

The only way I can explain that linear light makes the dark regions brighter is if the linear gamma build used a gamma lower than 1, and I'm pretty sure this would have been noticeable.

[Ver Greeneyes]

Originally Posted by 6233638

Comparing Bacondither's images, gamma definitely seems closest to the 16-bit source. [...] (neither are close in the upper range, but below about 50% gamma seems better)

It seems odd that the "16-bit source" image seems brighter than the others. If it was as dark as the others in the upper range, would linear be closer to matching?

Edit: Ah, bacondither mentioned this. The results are so close together that it seems kind of important.

Originally Posted by bacondither

The 16-bit source is a tiny bit brighter becouse i had to process it in another program(krita) that had support for >16-bit images and prosessing.
Kritas level controls did behave a bit strange.

[mithra66]

With Pioneer LX6090 display , gamma at 2.45, I vote for Gamma Light. Though I expected the opposite.

I don't know for sure if it's the most accurate, but I prefer its 3D effect, without the inconvenience of sharpening.

If LL wins the vote, I'd like GL to remain as some kind of "3D cool setting".

Anyway many thanks to Madshi -and all testers- to achieve this great work.

PS: DVD have never looked so nice thank to NNEDI3

[iSunrise]

madshi, I wanted to do some further testing with some TIF images (8bit/16bit per component), but I only get a black screen when loading them into madVR. LAV doesn´t seem to be the problem, because the file is being loaded and LAV also shows the file properties correctly, but the picture just stays black no matter what I do (disabled smooth motion to be sure).

Can I solve this somehow?

Here are two examples (first one 8bit, second one 16bit):
http://www.mediafire.com/download/6n7ji6q9f2hi2jp/BrightnessCal.rar
http://www.mediafire.com/download/9sbb5terv272meo/ColourRamp-1.rar

[sajara]

Originally Posted by 6233638

The "gamma light" image looks a lot closer to the "source 16-bit" image to me.
Linear light dither also has obvious banding, similar in width to each band in the "no dither" image.

I absolutely second this observation. GL is more smooth in 2 screens i tested. i can see quite obvious grey steps in LL which are attenuated and almost imperceptible in GL.

[Asmodian]

Originally Posted by bacondither

The 16-bit source is a tiny bit brighter becouse i had to process it in another program(krita) that had support for >16-bit images and prosessing.
Kritas level controls did behave a bit strange.

Dithering in linear light gives a more accurate image to the source.

I prefer LL over GL dithering and I do not think these test images can be used to decide given the above.

[Shiandow]

Originally Posted by Asmodian

I prefer LL over GL dithering and I do not think these test images can be used to decide given the above.

Actually, I tried to reconstruct what bacondither did but with the brightening performed in linear light, as far as I can tell this makes GL perform better (less banding, more accurate). I'll post some images in the near future, I still need to make sure that I haven't made any silly mistakes.

[Ver Greeneyes]

I did the same thing bacondither did but with device RGB spacing, and generating an expanded version of the original myself (it's my program, after all). There's no trickyness going on here: every image is expanded from 0-6 to 0-255, except the expanded original which I generated. Let me know if this helps clear up the difference between GL and LL.

no dithering
random dithering
gamma light error diffusion
linear light error diffusion
expanded original

(I think it shows that Gamma Light is more true to the source. How is the Linear Light conversion being done?)

[sajara]

Originally Posted by iSunrise

madshi, I wanted to do some further testing with some TIF images (8bit/16bit per component), but I only get a black screen when loading them into madVR. LAV doesn´t seem to be the problem, because the file is being loaded and LAV also shows the file properties correctly, but the picture just stays black no matter what I do (disabled smooth motion to be sure).

Can I solve this somehow?

Here are two examples (first one 8bit, second one 16bit):
http://www.mediafire.com/download/6n7ji6q9f2hi2jp/BrightnessCal.rar
http://www.mediafire.com/download/9sbb5terv272meo/ColourRamp-1.rar

opens here without problems

Originally Posted by Ver Greeneyes

(I think it shows that Gamma Light is more true to the source. How is the Linear Light conversion being done?)

Indeed it is. I also see that. Thanks for the original image.

[iSunrise]

Originally Posted by Ver Greeneyes

I did the same thing bacondither did but with device RGB spacing, and generating an expanded version of the original myself (it's my program, after all). There's no trickyness going on here: every image is expanded from 0-6 to 0-255, except the expanded original which I generated. Let me know if this helps clear up the difference between GL and LL.

no dithering
random dithering
gamma light error diffusion
linear light error diffusion
expanded original

(I think it shows that Gamma Light is more true to the source. How is the Linear Light conversion being done?)

I´m not sure how you come to that conclusion based on these examples.

If you compare your expanded original with the dark grays/blacks at the top left or bottom right edge of the gamma light example, it brightens the dark grays/blacks up quite a lot, while linear light has an extremely smooth transition down to black and finally reference black that closely resembles that of your expanded original.

Why would we suddenly accept something that brightens up all the important blacks by such a considerable amount? Black isn´t like the black in your original anymore, it´s a block of gray. This would seriously hurt dark movies or dark scenes.

Also, when you look closely, with gamma light I can see big repeating blocks of gray levels, while with linear light this is barely visible. The transitions themselves are way smoother with linear light.

This perfectly resembles bacondither´s findings, indeed. Because in his examples I also prefered linear light for both of these reasons.

Other than that though, for some reason the gamma light build gives the impression of a more dynamic and vibrant picture, sharper edges and it seems to be a bit better in terms of detail.

Quite frustrating, really. Because when looking at still pictures like that, linear light wins (IMHO), but when pictures start to move, that´s when our eyes seem to like gamma light more.

Originally Posted by sajara

opens here without problems

Thanks.

[Ver Greeneyes]

Originally Posted by iSunrise

If you compare your expanded original with the dark grays/blacks at the top left or bottom right edge of the gamma light example, it brightens the dark grays/blacks up quite a lot, while linear light has an extremely smooth transition down to black and finally reference black that closely resembles that of your expanded original.

Why would we suddenly accept something that brightens up all the important blacks by such a considerable amount? Black isn´t like the black in your original anymore, it´s a block of gray. This would seriously hurt dark movies or dark scenes.

I don't see this at all on my calibrated monitor. Both the expanded original and Gamma Light only go to black right near the end (matching a power law of x^(1/2.2), more or less), whereas Linear Light goes to black much sooner. It would be nicer to have a perceptual sample and perceptually expand everything so the transition isn't so sudden, but this will take a bit more work.

Originally Posted by iSunrise

Also, when you look closely, with gamma light I can see big repeating blocks of gray levels, while with linear light this is barely visible. The transitions themselves are way smoother with linear light.

I see them pretty clearly for both Gamma Light and Linear Light - but for the latter they stand out more near black, so I think Gamma Light looks smoother.

[iSunrise]

Originally Posted by Ver Greeneyes

I don't see this at all on my calibrated monitor. Both the expanded original and Gamma Light only go to black right near the end (matching a power law of x^(1/2.2), more or less), whereas Linear Light goes to black much sooner. It would be nicer to have a perceptual sample and perceptually expand everything so the transition isn't so sudden, but this will take a bit more work.

I see them pretty clearly for both Gamma Light and Linear Light - but for the latter they stand out more near black, so I think Gamma Light looks smoother.

Argh, I actually forgot to switch modes on my hardware calibrated monitor to a pure power gamma of 2.20 when I do these picture comparisons. That includes my last answer to bacondither, so everyone please ignore these posts. Sorry about that. You totally reminded me of that when you wrote that you also are on a pure power curve of 2.20.

Yes, you are perfectly right, gamma light is indeed more representative of the expanded original, because only the outer edge is indeed pure black.

At least for me, I think this is settled now, because I also strongly prefered gamma light when watching actual movies, anyway. My confusion suddenly has left the building. Hurray!

@madshi:
Just stumbled over a possible bug by accident. When I enable DCI-P3 calibration in the settings, I get this:


Going to bed now, way too late.

[cyberbeing]

Originally Posted by 6233638

I would much rather the algorithm be chosen based on objective tests for distortion/accuracy

I would as well, but madshi never set any such guidelines for testing such things. With it being a majority vote, the algorithm with the most preferred subjective appearance was always destined to win.

Originally Posted by leeperry

Nitpicking about grossly magnified and overexposed static screenshots is one thing but at the end of the day what matters is the subjective impression from a distance.

And this is why I decided to just take a step back and just watch from a distance which conclusions people came to through their own subjective testing. Here and there I'd give little tidbits from my objective tests to encourage closer examination in various regards, but that's about it. The irony is that way back in the beginning of testing, leeperry was strongly in favor of ED5 (floyd-stein, weight sum 1.00, old random generator), while at the end my top two favorites of NL5 (floyd-stein, 0.97, old random generator, bugfixes) & NL3 (floyd-stein, weight sum 1.00, old random generator, bugfixes) were probably closest to his initial preference. We both switched algorithms starting with the second build set.

Originally Posted by madshi

Modified the shader code a bit, but image output should be bit by bit identical to NL6.

Ah, in any case my testing on the linear and gamma builds was rather non-conclusive.

Originally Posted by Ver Greeneyes

linear light error diffusion

That image Ver Greeneyes posted of the linear light build expansion from 0-6 to 0-255 is rather concerning though, since there are 6 very distinct bands with multiple pixel wide noiseless areas on their borders. On the gamma light build there were also some much less distinct bands in the same areas that were slightly visible, but the transition between them was much smoother with smaller level jumps per pixel.

Originally Posted by iSunrise

@madshi:
Just stumbled over a possible bug by accident. When I enable DCI-P3 calibration in the settings

I have a similar bug-report regarding DCI-P3 and 3DLUTs here.