madVR - high quality video renderer (GPU assisted)

[coolcat]

Hi, I have a question, Advise me anybody please.
I use 60Hz monitor.
madVR Smooth Motion is off.
When I enter the Full Screen Exclusive Mode, fps is changed to 60fps with Smooth Motion off.
Is it a normal situation?
Full Screen Exclusive Mode always changes fps?
Please help me anybody,
Thanks a lot!

[Shiandow]

Originally Posted by cyberbeing

Video black brightened with Levels + Curves in Photoshop with my madVR 3DLUT active:
ED3 | ED5 | ED7 | ED9 | ED10

ED10 produces vertical line clumps on black.
ED9 has clumps.
ED7 is relatively uniform, yet most noise because of this.
ED5 has clumps.
ED3 is very low noise with clumps.

For some reason those images seem to be periodic when they should be random. If that is fixed somehow the 'clumps' and 'lines' you see should really only occur once. I fear that right now you're just comparing the amount of noise on 'black' and how 'lucky' the algorithms are at avoiding obvious patterns.

[cyberbeing]

Originally Posted by Shiandow

For some reason those images seem to be periodic when they should be random.

Since madVR is performing error diffusion dither on 16x16 blocks, I'd suspect that's is main reason why some of these periodic patterns form.

Originally Posted by Shiandow

I fear that right now you're just comparing the amount of noise on 'black' and how 'lucky' the algorithms are at avoiding obvious patterns.

I'm not. Those images in response to 6233638 were my first tests I've done on black. I honestly care more about how they behaved on the other 254 shades which are actually visible... All conclusions about the dither algorithms I've posted before then, do not take pattern behavior on black into account.

I also don't believe it is possible for any of these algorithms to be 'lucky', since madshi stated he designed the random elements to produce identical results on identical content so there wouldn't be any temporal noise. All the algorithms seem to behave rather consistently. Those which show a high percentage of patterns like ED9 & ED10, always have a high distribution of patterns wherever dithering occurs. On low motion content like anime, these patterns can be noticeable. With ED3, the quantity and size of the patterns was small enough to be almost undetectable. The fact that I could detect patterns at all with ED3 is why I bumped it down below patternless ED5 & ED7 in my anime test.

So far, my top two are ED7 (patternless, medium-low noise, very low crosstalk) & ED3 (low pattens, very low noise, low crosstalk) for the 'error-diffusion + random' slot.

ED5 (patternless, medium-low noise, low crosstalk) sits in a middle, as it's not very memorable being similar yet worse than ED7.

The worst two are ED9 & ED10 (medium patterns, low noise, medium crosstalk), with ED10 seemingly producing more 'strange' artifacts then ED9.
(ED1 I'm no longer testing since it's too noisy.)

[drew_afx]

are we comparing "quality" or PSNR?
the point of dithering is to make image "look" better at the cost of introducing noise...if I can't notice the noise in normal viewing conditions I'm all in for better color reproduction sharper image etc

[madshi]

Well, it should be said that the noise on black when using a 3dlut is not actually a "bug". The 3dlut black output seems to be a very very dark shade of gray. Almost black, but not completely. So proper error diffusion *should* throw in some stray dots. The reason why some algorithms do that more than others is that some of the algorithms don't spread 100% of the error to the surrounding pixels, but only ~97%. So if the gray is so near to black that the difference to black is smaller than those 3%, then there's a chance that there are no stray dots in the black, but it also depends on the exact weight configuration. I think we should mostly ignore this, and look for a different solution to the 3dlut problem.

So if we totally ignore stray dots on black, which algorithms do you prefer? And do you think I should only offer one algorithm? Or do you think I should offer two (one low noise, one higher noise)? Which two? FWIW, algorithms 9, 10 are somewhere between the low-noise and medium-noise algorithms, and 3 is somewhere between 9/10 and the medium-noise algorithms, in terms of noise level.

[cyberbeing]

Originally Posted by n3w813

I do use a 3DLUT in MadVR. When using the 3DLUT, my black point is raised just a bit from without a 3DLUT. So if I enable error diffusion (DC3 or DC5), I should be able to see the colored 'dots' on a 16-16-16 pattern right

Assuming you are in a pitch black room, then yes it should be possible to 'dots' with increased luminance when viewed in very close proximity, but it largely depends on your vision, and how well your display behaves near black. If near-black is being crushed with too little separation, then only a hardware instrument will likely be able to detect the changes.

As a simple test, can you see the single color channel noise on these images with your display?

Red (R1-G0-B0) uniform noise over black

Green (R0-G1-B0) uniform noise over black

Blue (R0-G0-B1) uniform noise over black

Originally Posted by drew_afx

are we comparing "quality" or PSNR?
the point of dithering is to make image "look" better at the cost of introducing noise...if I can't notice the noise in normal viewing conditions I'm all in for better color reproduction sharper image etc

Well this is where subjective opinion about what is and isn't visible clash.

I'm one of those people who always notice dither patterns, while minor +/- 0.33% low level random noise is undetectable during playback. Considering I've never seen a dithering pattern on an untouched HDTV broadcast or Blu-ray release, I'd assume most professional post-production studios have come to a similar conclusion and also favor patternless dithers at the expense of noise and accuracy. IMHO, a dither needs to be impossible to identify from random noise to achieve peak subjective image quality.

The other view in this thread is from those who do not notice most dither patterns, and prefer noiseless with minimum error and peak accuracy via PSNR or similar metrics.

Originally Posted by madshi

I am considering offering two different error diffusion variants: One with very low noise but with worm artifacts. And one without worm artifacts but slightly higher noise. So feel free to nominate either one or two builds. If you nominate two builds, one should be low-noise with worm artifacts, and one should be slightly higher noise without worm artifacts.

With madshi considering offering two variants in madVR, there is no reason why both views cannot be served. Though maybe we need three variants to make everyone happy...

Completely Patternless

Patterns Artifacts only

Patterns Artifacts + Worms

I'm of the opinion if we are having only two choices, one should be 'completely patternless' or we are just trading one type of artifact for another, considering 'worm' artifacts are just a side-effect of pattern layout. Similarly, the 'completely patternless' dither should have distortion less than 'random dither' or much of the benefit goes away.

[madshi]

Originally Posted by bacondither

Screenshots from big buck bunny with the brightness severly lowered in madvr to push many pixels to the low end of the 8 bit spectrum and then i brighten the image to reveal the dither noise in a image more realistic then solid grey bars...

A lot of pictures of a bunny with different dithering methods

Build 10 looks good but with a minimal grid pattern. Build 7 is free from artifacts and have a slightly higer noise level, but still it is lightyears lower then fully random dither.

That looks like a good test. I wonder if this is a fair representation of a real world test? Maybe we need 2-3 more comparisons like this with real world material to draw a final conclusion. Looking at the bunny images I like build 7 quite a lot.

[cyberbeing]

Originally Posted by madshi

So if we totally ignore stray dots on black, which algorithms do you prefer?

ED7, since it's patternless, sharp, and low distortion.

Originally Posted by madshi

And do you think I should only offer one algorithm?

No, since it seems other people want a completely non-random noiseless dither with patterns. May as well offer such an option.

Originally Posted by madshi

Or do you think I should offer two (one low noise, one higher noise)? Which two? FWIW, algorithms 9, 10 are somewhere between the low-noise and medium-noise algorithms, and 3 is somewhere between 9/10 and the medium-noise algorithms, in terms of noise level.

ED7, since it's patternless, sharp, and low distortion.

ED3, since it has minimal patterns, less noise than ED7, and appears to have the smoothest result of all the randoms.

While I do find it slightly worse, ED5 is similar enough to ED7 that it could be substituted in the patternless category. Beyond that, both ED9 & ED10 have around 2-3x the pattern size as ED3 which I find unacceptable. The non-random ones are obviously much worse, but as long as I have one choice I find acceptable, I could just ignore the second option.

[James Freeman]

Originally Posted by madshi

And do you think I should only offer one algorithm?

I strongly agree.

x1 non-random.
x2 quality random.

build 7 & 5 looks nice.

[Plutotype]

Originally Posted by bacondither

Screenshots from big buck bunny with the brightness severly lowered in madvr to push many pixels to the low end of the 8 bit spectrum and then i brighten the image to reveal the dither noise in a image more realistic then solid grey bars...

A lot of pictures of a bunny with different dithering methods

Build 10 looks good but with a minimal grid pattern. Build 7 is free from artifacts and have a slightly higer noise level, but still it is lightyears lower then fully random dither.

Thanks for those images. To me, all of the algorithms are still too agressive. Among these, case 5/7 for me, but I would still like to see a more grainy version, where black to gray transitions could appear smoother to human eyes. Optimum should be on a point, where noise will be minimized on blacker parts of the image and gradient transitions will be smooth - no apparent banding will occur ( I dont care about little noising here ).

[cyberbeing]

Originally Posted by Plutotype

Optimum should be on a point, where noise will be minimized on blacker parts of the image and gradient transitions will be smooth - no apparent banding will occur ( I dont care about little noising here ).

At a certain point you go beyond preserving lost bitdepth, and begin to enter debanding territory. If you enable madVR debanding, you may achieve the result you desire, assuming madshi has it as the very last processing step before dithering down to 8bit with error diffusion.

[bacondither]

Madshi you said that you are using alternating direction inside every 16x16 block.
Mabe using a method like in http://www.idav.ucdavis.edu/func/return_pdf?pub_id=1049 would help to eliminate these 16x16 blocky clusters of dots.

"Block interlaced pinwheel error diffusion with spiral or serpentine scan path" and

"The scan path goes outward for gray blocks, and inward for white
blocks. Note that during the processing of gray blocks, we diffuse errors from the gray blocks to the white blocks across block boundaries. To process each pixel, we first threshold the pixel value, then diffuse the quantizer error to its neighbors according to an error-weighting matrix..."

[Skankee]

9 and 10:
i still prefer 4 and 2 because it is more smooth at my test-sample.


@bacondither:

I cant reproduce your BBB-screenshots with those strong pattern, what did you change ?

When i look at Sample 6 (crf18@1080p) i cant see any pattern, and with enhanced brightness i can only see some minor pattern.
(Maybe this is source-dependent/better to see on low quality videos)

my untouched sample:http://s14.directupload.net/images/140209/jj8hgjtk.png
your sample:http://s7.directupload.net/images/140209/tled7muo.png

Maybe the tests should focus more on real world content. I cant see any pattern on my own samples without enhancing something, maybe someone can upload untouched samples where you can see pattern/worm artifacts ?

[NicolasRobidoux]

Originally Posted by 6233638

Well, it removes the obvious effects of ringing, but not all of them.

Using an old example, comparing Lanczos 3 AR with Lanczos 8 AR:


You can see that the right of the red circle is still affected, even though the obvious signs of ringing disappear.

This is a really good example of how the most reasonable yet simple overall approach to AR (which I assume Mathias followed) will sometimes be fooled. I expect that if I programmed my own, the first pass would be "fooled" in exactly the same types of situations. Thank you, as often, for your informative posts.

-----

Crazy month at work with the release of the IQ250 camera http://www.phaseone.com/.

Will take me a while to catch up with the questions I raised myself. But I'll get there eventually.

[Shiandow]

Originally Posted by cyberbeing

Since madVR is performing error diffusion dither on 16x16 blocks, I'd suspect that's is main reason why some of these periodic patterns form.

They don't seem to repeat every 16x16 block. I suspect that it is more a limitation of the random number generator, for some reason the random number generator just repeats itself. This also seems to happen on non-black colours, in my case it seems to repeat every 224 pixels. I strongly suspect that the algorithm somehow uses the position as seed for the random number generator and this seems to cause it to repeat itself when presented with identical content.

[Shiandow]

Originally Posted by bacondither

Madshi you said that you are using alternating direction inside every 16x16 block.
Mabe using a method like in http://www.idav.ucdavis.edu/func/return_pdf?pub_id=1049 would help to eliminate these 16x16 blocky clusters of dots.

"Block interlaced pinwheel error diffusion with spiral or serpentine scan path" and

"The scan path goes outward for gray blocks, and inward for white
blocks. Note that during the processing of gray blocks, we diffuse errors from the gray blocks to the white blocks across block boundaries. To process each pixel, we first threshold the pixel value, then diffuse the quantizer error to its neighbors according to an error-weighting matrix..."

That would be a nice addition, but it seems like it would be hard to actually implement. Not to mention that using that method on the same 16x16 blocks would use half the amount of threads so this could make it twice as slow. Although if this allows a smaller block size then it might actually be faster. Any block size from 10x10 or lower should allow more threads to work in parallel.

[leeperry]

So here I am after some good sleep and a full REC709 recalibration, I wanted to postpone my final answer as I wasn't sure where 5 would stand exactly...differences take a lot of concentration and are easier to spot with fresh eyes. And my previous XP calibration was a major failure on W7, I had to rework it all duh..

I understand you're willing to keep one odd and one even build, I find 2 technically impressive but too "flat looking" to my taste, on a LCD it doesn't look very natural to me...not like an "open window" as HD should. This would be great for DLP or Plasma, though.

7 is interesting but it's too fuzzy, 5 is the perfect in-between the sterile picture of 2 and the noisy 7....It looks very natural and highly enjoyable. I dearly hope that 5 will be nomitated as I really like the look



PS: I did try the 10 different builds.

[Seedmanc]

So has the nnedi thing been fixed yet? Doesn't seem to work on my GTX 560.

[6233638]

Originally Posted by bacondither

Screenshots from big buck bunny with the brightness severly lowered in madvr to push many pixels to the low end of the 8 bit spectrum and then i brighten the image to reveal the dither noise in a image more realistic then solid grey bars...

A lot of pictures of a bunny with different dithering methods

Build 10 looks good but with a minimal grid pattern. Build 7 is free from artifacts and have a slightly higer noise level, but still it is lightyears lower then fully random dither.

In real-world images like this, it does seem that 7 is producing better results than anything else, and the noise does not seem to be that much higher.

Originally Posted by Plutotype

Thanks for those images. To me, all of the algorithms are still too agressive. Among these, case 5/7 for me, but I would still like to see a more grainy version, where black to gray transitions could appear smoother to human eyes. Optimum should be on a point, where noise will be minimized on blacker parts of the image and gradient transitions will be smooth - no apparent banding will occur ( I dont care about little noising here ).

Keep in mind that these examples have been brightened significantly, and that the goal of dithering is not to eliminate banding in the source, but to convert 16-bit to 8-bit data.

[GREG1292]

Build 7 is best for my projector and has plenty of pop plus a very clean picture. Don't know if I can improve on my settings but I use needi3 64
and dxva2 for the other 2. Gpu and CPU Q are 24/32 all quality settings unticked and doubling 256 for both luma and chroma on a gtx770 with nvidia 327.43. Picture is amazing with no frame drops. Might look a little
over the top but that's the way I like it. Same settings on build 2 not
playable. Custom fan curve gpu never gets hotter than 60 Celsius.
My vote so far is for 7 and all I can say is wow!!!!!!!!!! Thanks for the awesome work 720p and 1080p movies are so real looking and no black
artifacts or noise seen in the image. Using a Mitsubishi Hc7900 on a 92"
screen sitting 12 ft back and gives the feeling of being inside the screen.
So I would say that build 7 is my favorite so far for a movie watcher enthusiast.


Sent from my iPhone using Tapatalk