madVR - high quality video renderer (GPU assisted)

[*Touche*]

Yes, this effect is what I earlier called the "dirty screen effect". You're describing it as a "dirty lens effect", which is basically the same thing. I do wonder, though, whether this effect is visible/distracting at all at 8bit. If it's not, maybe it'd be better to stick to the static builds to keep the subjective noise as low as possible? I'm on the fence on this one. It's definitely a trade between lower subjective noise floor (pro static) and dirty screen/lens effect (contra static).

I wote NO on static. While it is much less obvious in 8bit, I can still see a benefit with dynamic builds on moving objects.

@leeperry, do you like ClearType? I think that's somewhat similar to Multi/OppositeColor. I've read that some people are bothered by ClearType and prefer it turned off, while the majority of people prefer it turned on.

I like oppositeColor, but I hate ClearType with a passion :)

[James Freeman]

In the 4-bit builds where the effects of either are clear, dynamic is less harmful to the image than static.
Static dither patterns are very obvious, dynamic adds noise, but blends in with the image more.

I agree that the dynamic build looks smoother overall in 4-bit.
May be I should let go of thinking the Static build will be "good enough" in 8-bit, if we are aiming for best looking dithering (even if it is below subjective human vision, or right on the threshold).

Maybe "Good Dithering" IS good amount of noise to our eyes?

One more trick,
Try to lower the Saturation in MPC to lower the visible Color influence, and test the builds again.
The results may vary...


Also,
You fellows who judging these build on a TV's,
Sharpness setting will influence your judgement big time.
Chroma subsampling of the panel (4:4:4, 4:2:2, 4:2:0) will make difference in dithering size and how it looks.
Take that into account.

[bacondither]

The temporal tries didn't work out so well for now, unfortunately.

I used a poor choice of words, what i meant was the temporal smoothing effect when using different randomisation seeds every frame does like in OppositeColor dynamic and multiColor dynamic builds.

I agree that the "temporal" builds using cyberbeings idea did not look so good.

[wolfman2791]

Maybe we should take the best out of the last 3 batches and then compare them?
- Adaptive 4
- The choice of the monocolor/multicolor and static/dynamic batch
- The oppositic color choice

I think there may be too many difference builds floating around and it may be getting unmanageable.

[bacondither]

Maybe we should take the best out of the last 3 batches and then compare them?
- Adaptive 4
- The choice of the monocolor/multicolor and static/dynamic batch
- The oppositic color choice

I think there may be too many difference builds floating around and it may be getting unmanageable.

The Adaptive 4 build is the same as the monocolor static build.

[wolfman2791]

The Adaptive 4 build is the same as the monocolor static build. Lol, you know, the difference between some of these is so slight that i can't even tell... i'm using the naked eye and not testing algorithms.

I think i'm going to shut my unknowledgeable mouth now :):D

[mithra66]

And I'm finally back to madVRadaptive4, I feel like I'm watching through to the TV instead of at the TV itself, my eyes instantly feel at home and can focus on that good ole monochrome movie grain, all those color builds feel like a bad dream and I'm stunned all over again

I can't seem to go back to A4 as to me it looks like something is missing? I'm so far into the picture with ODC build
I don't need to concentrate on the grain but the movie it's self. A4 i'm alway's trying to make the picture look better.
Very distracting indeed.

I'm confused. Didn't you say you prefered oppositeColor_dynamic? So you changed your mind again and now you prefer adaptive4 again? At least your first paragraph reads like that. But then in your 2nd paragraph you're saying adaptive4 looks like there's something missing, which is a negative judgement? So I'm not sure right now whether you like adaptive4 or not.



I think there is a missing quote from Lepeery for the first sentence. SO Greg would still be for ODC.

I have a better feeling when watching Opposite Color, this seems more realistic to me. I don't dislike A4, but it seems to me that Opposite Color to A4 is similar to what A4 is to no dithering, although with smaller difference: artificial sharpness and dynamic versus fineness.

I observed something odd on 1-25 low luminance grey bars: I have some color dithering no matter if Madvr's dithering is activated or not. Which is expected as my display is a plasma.
However I have some obvious (from 30cms distance viewing) green dithering starting at 24 with MonoColor and it is starting at 25th level for OppositeColorDynamic. Maybe some noise cancellation phenomenon?

Finally my PQ appreciation would be:
No dithering: 90/100
MonoColorStatic or A4: 96/100
MultiColorDynamic :96.5/100
OppositeColorDynamic: 98/100
Future MadVr improvements: 99.999/100 :D

I prefer OCD but I think I could survive with A4. ;)

[annovif]

Madshi, I tried carefully on bluray jinc3AR against the new Needi3 256 neurons. Incredible!! Improved quality with superior sharpness. Not to mention the rest, we're pretty much ready for the 4k at the highest quality imaginable. I think that there are no words to express my gratitude for the work you are doing and that the community can not do without your genius. This is also why I decided to give yourself A SMALL AMOUNT OF MONEY AND TO INVITE EVERYONE TO DO THE SAME. For years, you are at the service of all of us putting time, patience and effort but it is time that we do something for you. Thank you, thank you, thank you.

Ps: Ehhmmmm, where is the "DONATE" button ?

[wolfman2791]

I agree... MultiColorDynamic and OppositeColorDynamic are the top 2 for me as well. Naked eye test. Nothing scientific.

[huhn]

Madshi, I tried carefully on bluray jinc3AR against the new Needi3 256 neurons. Incredible!! Improved quality with superior sharpness, but with almost the same gpu consumption. Not to mention the rest, we're pretty much ready for the 4k at the highest quality imaginable. I think that there are no words to express my gratitude for the work you are doing and that the community can not do without your genius. This is also why I decided to give yourself A SMALL AMOUNT OF MONEY AND TO INVITE EVERYONE TO DO THE SAME. For years, you are at the service of all of us putting time, patience and effort but it is time that we do something for you. Thank you, thank you, thank you.

Ps: Ehhmmmm, where is the "DONATE" button ?


is there a possibility you are an nvidia user? it's only working with old driver with nvidia.

why did i say this? very simple the gpu usage with needi 256 is so much higher than jinc 3 ar that it doesn't make sense to say it is nearly as fast as jinc 3 ar. i think there is no gpu out there to do 1080p to 4k with needi >256< neurons even 540p needi 256 30 fps needs a really fast gpu. but jinc 3 ar is like idle compared to this.

so are you really sure?

[annovif]

I'm sorry but i'm an Ati 7750 user and my pj is only 1080p . I was talking about the chroma setting of a bluray 1080p too.

[Buckster]

I'm surprised with a 7750

I'm running an overclocked 7850 and can only manage 32 Neurons to get no dropped frames, anything 64 and above is lag-fest

[Thunderbolt8]

I dont really have a clue about all those test patterns and stuff, but Id suggest the following anyway:

- since madVR is used as a video renderer, comparisons should almost exclusively be made with videos and not some kind of grey ramps. its rather pointless to use something else as comparison for which madVR is not meant to be used.

- it would also help if madshi could state each time what to look out for when comparing stuff, which factors and which others can or should be disregarded. e.g. in case of error diffusion low noise smoothness is what should be achieved and not crispness. so the factor crispness should be disregarded when doing comparisons no matter how good it might subjectively look or some kind of pop effect or whatever. imho its vital that people stick to this when doing their comparions, but also that madshi points out such things each time this should apply. otherwise all this testing wont lead to anything.

[GREG1292]

OCD for me as a4 has more pop but OCD is more detailed and accurate and a deeper and more detailed depth going back into
the screen. More natural or balanced. All are good and if we
have access to all the test builds
In one folder we can pick and choose. Problem solved for everyone with only one build needed. Thanks everyone as we
all have different eyes and viewing.

We need to all use the same setting as madshi if he would share
so we can give input that is similar. I know screens and video cards
are different but if we use default for MadVR and our graphic cards
would that not level the playing field? We need a standardize test
as not everyone has the same gpu power. Since I am not a programmer I am just going to leave my input as OCD or R2D2 and a brief description.

You all know that I prefer the GTX770 over the 7870xt
on my projector and that is with everything default on both
madVR and gpu settings. Where the GTX770 pulls ahead is when
gpu load increases and remains smooth and stutter free with no
over heating with neurons used.

Sorry for the additional comments I like testing but sometimes feel it is for
my own selfish needs.

[James Freeman]

Try to lower the Saturation in MPC, or remove it completely (-100) to lower/remove the visible Color influence, and test the builds again.
The results may vary...


Quoting myself because I want to emphasize this.

With no color at all (Saturation to -100).

I can see colored pixels the MultiColor dithering creates in 4-bit with real content.
Then I quickly changed the desktop folder name to something else to return to 8-bit (no need to close the paused video).
Now I can't see any color what so ever.

To verify that the colored dithering is still there in 8-bit (although completely unseen now), I took some screen shots of 4-bit & 8-bit (same frame) and boosted the Saturation in photoshop.
Yes, the 8-bit has the same colored dithering but I needed to boost the saturation four times more compared to 4-bit, to see the same level of saturation.

Still,
OppositeColor Dynamic is absolutely the smoothest looking, in 4-bit or not.


P.S
On 4:2:2 TV's the colored dithering pixel will be TWICE as big horizontally, that may bias some users to falsely judge the Multi/Opposite Color builds.
So please try to test the builds on a 4:4:4 monitor or TV.

[Shiandow]

I think so, too. At least I see no better solution right now.


The only convincing alternative I can think of is the one I posted a few pages back where you simply pick a random color (r,g,b) such that r+g+b=0. The Yuma noise from this algorithm should be significantly lower than the one from the opposite algorithm, all channels have the same distribution, and all values between -0.5 and 0.5 are possible. The only downside I've found is that not all values between -0.5 and 0.5 are equally likely, values around 0 are twice as likely as those at the extremes (this is because there are twice as many possible values for g,b such that r+g+b=0 if r is near 0). This effect may actually not be a bad thing but you'd need to check that gradients still look nice and possibly retune the algorithm to account for the different noise distribution.


Yes, this effect is what I earlier called the "dirty screen effect". You're describing it as a "dirty lens effect", which is basically the same thing. I do wonder, though, whether this effect is visible/distracting at all at 8bit. If it's not, maybe it'd be better to stick to the static builds to keep the subjective noise as low as possible? I'm on the fence on this one. It's definitely a trade between lower subjective noise floor (pro static) and dirty screen/lens effect (contra static).

FWIW I think that I can occasionally see dirt on my screen that causes less than a 1/256th difference in brightness. Although I don't think I'd be able to tell the difference between this and 'artificial dirt'. On the other hand if my screen was kept meticulously clean then I might very well be able to notice 'artificial dirt'.

Edit: I actually made a mistake when measuring the amount of Y noise, the oppositeColor method seems to be better than the alternative I proposed. I thought it was weird that there didn't seem to be much of a difference between oppositeColor and multiColor was so small, turns out that the numbers I used for oppositeColor weren't random. Anyway I now have a mathematical 'argument' that oppositeColor causes less Y noise than multiColor.

[leeperry]

I might be mistaking but some of you would appear to base their opinion on the 4bit trick? Nev has made quite clear that this doesn't quite scale with 8bit...maybe madshi could provide a way to go 6bit? It would be closer to 8bit and might actually be useful to the 6bit TN crowd(FWIR mVR doesn't currently officially support ED for 6bit).

There's a very good chance that viewing in 4:2:2 will be affecting the results. I do all my testing in 4:4:4.
I clearly remember that a while ago you were advising to go 4:2:2 if the TV would then allow noticeably better post-processing and quite frankly I'm not trading 24/50Hz support & BFI for 60Hz-only 4:4:4. Not a good deal at all as far as movies are concerned, OTOH computer fonts eventually look heaps better.

Also, several Sammy 1080p TV reviews claim that if you disable BFI you only get 360p..

How do you feel "adaptive4" compares to "mono static" ? (not mono static temporal)
I assumed that these two builds were identical from madshi's explaining, maybe he could confirm?

I have just done some more testing with the "static" builds, and I know what it is that I don't like about them.

When the camera moves, e.g. in a hand-held shot where there is a subtle swaying motion, because the dither pattern is static it looks like there is something on the lens, as it moves with the camera.

With the dynamic builds, it's randomized so this does not happen.
Yes the noise level is somewhat higher, but I just don't see it being an issue when viewing at 8-bit.
Yes, this effect is what I earlier called the "dirty screen effect". You're describing it as a "dirty lens effect", which is basically the same thing. I do wonder, though, whether this effect is visible/distracting at all at 8bit. If it's not, maybe it'd be better to stick to the static builds to keep the subjective noise as low as possible? I'm on the fence on this one. It's definitely a trade between lower subjective noise floor (pro static) and dirty screen/lens effect (contra static).
I am not seeing any dirty lens or DSE whatsoever, maybe BFI is saving the day on this one? OTOH with dynamic builds, the dither grain is way too random & shaky and kills the original movie grain appearance to my eyes.

Let me add the original image to make the comparison more helpful:

| -- monoColor (http://madshi.net/monoColor.png) -- | -- multiColor (http://madshi.net/multiColor.png) -- | -- oppositeColor (http://madshi.net/oppositeColor.png) -- | -- original (http://madshi.net/edTestOrg.png) -- |

As you can see, there's no grain in the original image, and the background is supposed to be blurred. Ok, so this test image does not help us decide which of the error diffusion algorithms reproduces film grain in the best possible way. We need a different video/image for that. But this image does show us which error diffusion algorithm produces the lowest subjective luma noise levels. Agreed?
Well, you are taking a 8bit source, processing it in all kinds of ways in 16bit and then outputting to 8bit again....there will obviously be some resolution lost? And this is the mid-noise algorithm, I'm sure you could lower it and/or possibly use other values than 1.0 or 0.97 weight sums(still very curious about what that'd do :D).

do you like ClearType? I think that's somewhat similar to Multi/OppositeColor. I've read that some people are bothered by ClearType and prefer it turned off, while the majority of people prefer it turned on.
I am forced to run Windows in 4:2:2 and computer fonts look pretty nasty (http://s807.photobucket.com/user/floatpoo/media/444_faq/test2_c.png.html)...CC makes it even worse and I tried every flavor using the W7 CC tuner.

OCD for me as a4 has more pop but OCD is more detailed and accurate and a deeper and more detailed depth going back into the screen. More natural or balanced. All are good and if we have access to all the test builds In one folder we can pick and choose. Problem solved for everyone with only one build needed.
Good to know, so if your brain can stand the chroma rainbow noise with no sweat then it's a good bet I guess :)

If it's a deal breaker or if all you care for is pop "über alles" then A4 is a good choice too IME/HO.

Hopefully at some point madshi with decide to go with a default algorithm and allow the minorities to pick their favorite using a registry key...if providing that choice in the UI is truly out of the question.

Having a higher luma noise floor can make the image appear more detailed. But that's most definitely *not* the purpose of error diffusion. The main purpose of error diffusion is to simulate a higher bitdepth as much as possible. So the better the error diffusion works, the less crisp the image might actually appear, compared to a higher-noise alternative. So what we should be aiming for is not crispness, but low noise smoothness. If you want that small amount of added crispness, you can add artificial grain, or sharpen the image
Fine, but A4 does it all so well...all roads lead to Rome you know. Sharpening usually goes along with nasty white halos and artificial grain always appears to be living a life of its own...only GrainFactory3() would allow to choose the strength and chunk size for dark/mid/bright pixels but even with a lot of finetuning it still didn't look as "natural" as A4 and it's a major CPU hog too.

So does that means that the softer look & motion blur of the color builds is by design and more "technically optimal"?

All this said, I come from the CRT/DLP world and I basically want my LCD to look like them. Many ppl claim that DLP comes with a stunning "cinema-like" analog look and I was strongly impressed by my old 3K:1 Mitsubishi HC3100 (http://cineforhome.com/reviews/projectors/MitsuHC3100-1100/HC3100-1100Review.htm) on a daily basis using a 2m wide projection screen from a 3.3m distance. It looked so real and so enjoyable. PQ was a real thrill thanks to the 600:1 ANSI contrast and invisible black during bright scenes(with its iris closed).

DLP looks very sharp thanks to the single panel(and mineral glass lenses on the HC3100) but also grainy due to the mirrors holders (http://www.hometheaternetwork.com/pics/DLPchip.jpg) and it's a known trick to use grain in order to deblock.

Also, FWIR even the most optimal bluray 8bit encodes are always a compromise when it comes to movie grain...all this to say that subjectively speaking A4 does it all at once for me: deblocking, natural looking grain-based edge enhancement, analog look, reinforcing movie grain without changing its pattern/direction, etc etc..

A4 also saves the day with upscales and I still believe that every display comes with its own dithering pattern and it's obvious that we all more or less see differently so there can't be such a thing as a magical "one size fits all" mVR dithering algorithm IMHO. Also, even with BFI LCD is still not nearly as responsive as CRT.

I wish I still had a bigger screen so I could do my own testing.
IIRC Joe Kane made clear in that very interesting video (https://www.youtube.com/watch?v=0ZqhA3iIHm4) that you need as big of a screen as possible in order to seriously evaluate video processing.

I find 32" from a 80 cm distance to be perfect for 1080p content, but it'll sure take something seriously big to start whining all over again about 4K dithering to madshi :D

I to am only concerned about the picture
Thank you for the kind words, I learned a lot from ppl like MLill on HCFR (http://www.homecinema-fr.com/forum/membre/MLill/).

[*Touche*]

It looked so real and so enjoyable. PQ was a real thrill thanks to the 600:1 ANSI contrast and invisible black during bright scenes(with its iris closed).

That doesn't look impressive.

[Soukyuu]

True. But XP support is only one out of several reasons why I'm using D3D9.Does anything (besides the porting effort) speaks against switching to openGL+openCL and go multiplatform? I think you mentioned going pure openCL at some point.

[leeperry]

That doesn't look impressive.
Well, +500:1 ANSI for a video projector is quite high AFAIK...at least it was when I was using it in 2007. And I wasn't dressed like a ninja in a room with black walls ^^

MLill kept a measured specs comparison here (http://www.homecinema-fr.com/forum/videoprojecteurs-hd-ready/tableau-comparatif-entre-projecteurs-ii-t29688637.html) and you can see that even in 2012 there are few projectors that would provide >500:1 measured ANSI contrast, most of them being DLP for that matter.

I of course thoroughly enjoy how on a LCD flat screen the ANSI contrast is equal to the native contrast :)

[nevcairiel]

Does anything (besides the porting effort) speaks against switching to openGL+openCL and go multiplatform? I think you mentioned going pure openCL at some point.

OpenGL makes native DXVA impossible (or at least pointless), and also doesn't give you access to DXVA deinterlacing or DXVA scaling anymore.

So a feature loss, for what gain?
Its not like other platforms have DirectShow, or even any comparable video frameworks where a renderer could just "plug in" like it can in DirectShow.

Not to mention that Error Diffusion uses DirectCompute now, which is part of DirectX, and a Windows technology. ;)

[DragonQ]

I've never understood the love for projectors. Sure it's a relatively cheap way to get huge pictures but it's impossible to get as good black levels as most LCDs, let alone plasmas, even in dark rooms. I remember watching clips of LotR at a friend's house a few years ago on his projector and I wasn't impressed at all.

[Buckster]

currently comparing the image output of the various experimental builds- but have to echo other people's views in that its great to even be able to try out different flavours of picture processing - especially as some is personal taste really

with the new processing, OpenCL upscaling etc - I've not seen a better image output to my Panasonic Plasma (Pro PF11) - and therefore would like to express my great thanks to Madshi

[6233638]

I clearly remember that a while ago you were advising to go 4:2:2 if the TV would then allow noticeably better post-processing and quite frankly I'm not trading 24/50Hz support & BFI for 60Hz-only 4:4:4. Not a good deal at all as far as movies are concerned, OTOH computer fonts eventually look heaps better.Yes, for viewing it's worth sacrificing chroma resolution if it means you gain better image quality in other areas, such as backlight scanning on an LCD.

But if you're trying to critically judge something like dither algorithms where pixel-level details matter, I wouldn't do it in anything less than full resolution.

I am not seeing any dirty lens or DSE whatsoever, maybe BFI is saving the day on this one? OTOH with dynamic builds, the dither grain is way too random & shaky and kills the original movie grain appearance to my eyes.Dithering is much finer than film grain. In fact, I have yet to find any sources where it has a visible effect in 8-bit.

Every time I think I am seeing noise in the image from dithering, it turns out to be noise that is encoded in the source. (just as visible when I disable dithering)

In fact, I am finding that due to the noise level being lower with error diffusion rather than random dithering, that sort of thing is a lot more noticeable now.

Good to know, so if your brain can stand the chroma rainbow noise with no sweat then it's a good bet I guess :)So do you find the color and opposite builds to be noisier in this image (http://abload.de/img/ditheringffu1i.gif)?

All this to say that DLP looks very sharp thanks to the single panel(and mineral glass lenses on the HC3100) but also grainy due to the mirrors holders (http://www.hometheaternetwork.com/pics/DLPchip.jpg) and it's a known trick to use grain in order to deblock.Just use the 4-bit mode all the time if you want a more DLP or Plasma-like image. The much-revered Kuros only use 14 shades when drawing an image, and 4-bit gives you 16...

[iSunrise]

And it's incorrect that you say that we add more noise. That's simply not true. If you look at real video content, the oppositeColor build has the lowest subjective noise level.
I guess that´s where our wording differs a bit and what we mean by it, so I´m trying to make it more clear. I´m not even talking about subjective noise floor/level or subjective noise perception, because these can clearly be fooled easily, as can our perception of depth (crisper image = better image). Because I actually agree, ED4 clearly gives a crisper image (not necessarily a good thing, but since you asked about subjective analysis, that´s what we did).

However, the images that were provided (e.x. by you and 6233638) clearly show that more noise is added, not less. There´s more noise dots and they also are closer together, meaning, it´s way thicker than with the ED4/mono_static build (that´s the smoothness you´re talking about), so basically, it´s filling way more gaps as before with the ED4/mono_static build. And since we are only comparing still shots (where this is already visible), when being in motion, there´s even more noise which is changing it´s pixel position all the time, because the noise is dynamically changing, whereas it was static before. That doesn´t have to be a bad thing and I didn´t say as much. Actually I already said the opposite myself, before I even did a subjective test, because when I am thinking about changing picture contents, the noise should adapt to the contents or else, we wouldn´t technically be dithering anymore, we would just add noise. The problem with this is, that on static parts of an image (where there are usually quite a lot of), you would have a lot of changing noise all the time. This was the trade-off we were talking about.

Now, concerning the noise itself. If it´s visible or not, is a completely different story. And that´s all I was trying to say. Now, the question is, what we want to achieve and I agree with you, we should strive for an accurate appearance.

But how do we know that we are accurate? Shouldn´t the dithering algorithm only add noise where it is needed? Why don´t we add even more noise that is closer together? And I was just thinking out loud, if we already crossed that bridge, yet.

Now, comparing the builds, while the multicolor_dynamic build also makes that additional noise stand out quite a bit (that´s a bad thing), the opposite_dynamic build reduces the visibility of the noise by quite a lot. So, yes, the opposite build clearly has a lower perceived noise floor compared to the multi_dynamic build. But ED4 still has the least amount of noise (just look at the sky picture, provided by 6233638), even though it´s clearly more visible. Which is the trade-off with ED4.

Since we are currently evaluating the medium-noise dithering algorithm (which the opposite_dynamic build comes closest currently), the aim should be "as accurate as possible, like RD with a lot less noise", while the aim for the low-noise algorithm should be "accurate, but only as much noise as needed". Or how would you define that yourself?

[Buckster]

for a Panasonic Plasma (PF11 with DualHDMI board) - should I be aiming to output 4:4:4 from my ATI card or 4:2:2 ? I thought my panel accepts RGB 4:4:4 - but I struggle to get to see any of the black levels in the black levels test

[leeperry]

if you're trying to critically judge something like dither algorithms where pixel-level details matter, I wouldn't do it in anything less than full resolution.
If I disable BFI I end up with 360p motion(from what many reviews are claiming at least, and indeed it's motion-blur galore to my eyes), pretty nasty flickering and I'm stuck in 60Hz-land, I don't see how that's gonna help with anything. Especially as YV12 is 4:2:0 to begin with.

Every time I think I am seeing noise in the image from dithering, it turns out to be noise that is encoded in the source. (just as visible when I disable dithering)
I see where you're going but that definitely doesn't match what I see in Oblivion as the "dynamic" mechanism utterly veils the picture to my eyes.

So do you find the color and opposite builds to be noisier.
All those static screenshots are undeniable, especially as madshi made clear that the goal is softness.....too bad those colorful builds look blurry in motion to me(and to a few others as well apparently).

Just use the 4-bit mode all the time if you want a more DLP or Plasma-like image.
The HC3100 is using 10bit processing with the DDP3020 chip: http://www.mitsubishielectric.com.au/354.htm
equipped with a combined full 10-bit processing I/P converter and scaler, and a 12-bit floating point digital gamma controller. The result is approximately four times the number of gradations of a conventional 8-bit model

[XMonarchY]

OppositeColor looks much cleaner than others to me. I don't know why my opinions differ so much from others, maybe its the 4:2:2 3000:1 CR SPVA that makes things worse with noisy builds or whatever else, but I like OppositeColor the opposite the most... I also like the least noisy ED build the most (Noisy2Limited I think is the name).

Should we not have a lot more people in this sample group to really figure out what the public likes the most?

[XMonarchY]

Also, I wonder if the broken OpenCL <-> D3D9 interlop is because nVidia does not feel like supporting D3D9 is very important. I think Windows XP is too ancient to support. Who is going to use ED and NNEDI3 on Windows XP??? You need a powerful card to run it and all powerful cards users have Windows 7 or Windows 8.

[leeperry]

Actually I don't think you can get OCL support under XP at all, at least it wasn't recognized as active by GPU-Z with a HD7850 and the latest AMD drivers a few months ago: https://bitcointalk.org/index.php?topic=117956.0
I have a 7870 running XP and have the same problem. No OpenCL devices. have tried all the different driver versions I could get my hands on, and the different SDK's to no avail.
And DirectCompute is part of DX11.

[annovif]

I'm surprised with a 7750

I'm running an overclocked 7850 and can only manage 32 Neurons to get no dropped frames, anything 64 and above is lag-fest

I had the old Ati driver and with windows 8 32 bit chroma settings Needi3 from 16 to 256 neurons i had 30% power consumption, instead with windows 7 i had 70% .

I've now the new Ati driver and the situation is not the same. The pc freeze when i select needi3 . Mah!!

Ps: I want to donate in any way:D

[mzso]

Hello madshi!

There's a crash with this file if I also use LAV video. Nevcairiel says (http://forum.doom9.org/showpost.php?p=1670015&postcount=17172) the crash is inside madVR. (v0.87.4)

https://drive.google.com/file/d/0ByfdfPvnoDuzSlZqY1JxeDdXNkU/edit?usp=sharing

[XMonarchY]

This may be a little off-topic, but I have an issue here related to the only 327.23 drivers that work with OpenCL...

nVidia control panel reports accurate and properly available refresh rates that my TV supports, but Windows 8.1 monitor settings window does not. The "Hide unsupported refresh rates" box is ticked, but it still shows 29hz and 50hz, neither of which are supported. This creates a big problem for some media playing software that attempts at using those refresh rates, which results in a black screen and a "Unsupported Mode" error from my TV... Does anyone else get unsupported refresh rates listed in Windows 7/8 monitor settings using 327.23 divers?

[Sievert]

Thanks God I won't be around for quantum dithering...that'll be fun. I bet randomness will come at no cost.

[Shiandow]

Thanks God I won't be around for quantum dithering...that'll be fun. I bet randomness will come at no cost.

I bet there will be some people who claim it looks better when unobserved.

[Guest]

Did some 4bit mode tests with Star Trek: Voyager on my IPS monitor, and Multi-Colored dynamic one is a clear winner to me:

Mono-Colored (http://i4.imageban.ru/out/2014/02/20/ff4af591dbab5fa262f2e3959c2effe2.png)
Multi-Colored (http://i2.imageban.ru/out/2014/02/20/32f5f0af33184c268e3984c80ff33cc7.png) Little mod bump because this post got caught in the mod queue. Welcome to the forum!

[ryrynz]

Multi-Colored dynamic one is a clear winner to me


I agree based on that pic.

[MistahBonzai]

I bet there will be some people who claim it looks better when unobserved.

Beautiful..just simply beautiful!

[iSunrise]

Since the 4bit mode was dismissed as not giving an accurate representation of things, I did one more final test without the 4bit mode, feeding madVR with a 16bit per component gradient (48bit total) that I downloaded from Lightillusion (http://www.lightillusion.com/). The original file I´m talking about is called BWRamp.tif.

The goal with this is to see if there´s any hint of visible coloring when e.x. we mainly have black/gray/white color tones (like e.x. B&W movies) and I also wanted to see how far we´ve come since we began with the ED algorithms (also in comparison to random dithering, since the 4bit mode doesn´t allow that either).

And since I don´t have the equipment to compare things like coloring when you look at moving images, movies or other clips, which are usually not synced, side by side and two displays with perfectly matching calibration I went with what I can do.

What I did:
(1) I took screenshots with no dithering, random dithering, the mono_dynamic and opposite_dynamic dithering builds
(2) I used GIMP to reduce the input range from 0-255 to 192-255 and the output range from 0-255 to 0-128 and applied the change
(3) In the next step I reduced the input color range from 0-255 to 0-32 and applied the change
(4) I opened every screenshot with Irfanview and cropped out the remaining range of the gradients

Here are the screenshots:

no_dithering vs. random_dithering vs. mono_dynamic vs. opposite_dynamic

http://abload.de/thumb/bwramp_no_dithering_go3kea.png (http://abload.de/image.php?img=bwramp_no_dithering_go3kea.png)http://abload.de/thumb/bwramp_random_ditheri4djlu.png (http://abload.de/image.php?img=bwramp_random_ditheri4djlu.png)http://abload.de/thumb/bwramp_mono_dynamic_gupj8z.png (http://abload.de/image.php?img=bwramp_mono_dynamic_gupj8z.png)http://abload.de/thumb/bwramp_opposite_dynam6jjdm.png (http://abload.de/image.php?img=bwramp_opposite_dynam6jjdm.png)

Now, what should be rather obvious is that without dithering, when there´s substantial processing involved, or if the bit-differences (input <-> output) are too high, you are left with only fragments of the original.

When going to the next screenshot, we have random dithering. While we clearly have preserved most of the steps, the result looks unnecessarily noisy and rather wild. This is why we wanted a more accurate dithering, with a lower noise floor and a smoother appearance.

Now, the interesting comparison.

If you look at mono_dynamic and opposite_dynamic from a distance, you will see that opposite_dynamic has an even smoother appearance than mono_dynamic. However, if you look closely (side by side) at both of them, you can see coloring in the opposite_dynamic shot. The amount of the coloring is high enough that on a calibrated monitor, my eyes can see it, side by side after one another.

(5) I opened both screenshots with Paint and took a measurement of 74,131 (just an example, I could pick other spots, same results) and compared the results:

http://abload.de/thumb/bwramp_paintfoagp.png (http://abload.de/image.php?img=bwramp_paintfoagp.png)

To make it easier, here are the colors I took:

mono_static vs. opposite_dynamic

http://abload.de/thumb/bwramp_mono_dynamic_cz9bt6.png (http://abload.de/image.php?img=bwramp_mono_dynamic_cz9bt6.png)http://abload.de/thumb/bwramp_opposite_dynamhrbyx.png (http://abload.de/image.php?img=bwramp_opposite_dynamhrbyx.png)

While mono_static gives a gray tone, where there should only be gray, opposite dynamic colors it with quite a lot of green. And it´s not just that one spot, most of the gradient is colored, that´s why I was able to see it. Whether this also translates to RGB instead of just gray tones (luma) I don´t know, I haven´t specifically tested that.

These problems will probably not show in such a drastic form, since we usually watch with 8bit input (+16bit processing) and finally, 6/8bit (+FRC) output. But they are clearly there.

So, when looking at movies, when it comes to pure smoothness, I would say the opposite_dynamic build wins here. When it comes to clarity and color accuracy, I still prefer the mono builds.

Hopefully madshi has a way of improving the color accuracy, then the choice should be easy. Maybe someone else has an explanation for my results, then I would like to know.

[*Touche*]

But what would the color be in the next frame in the same spot? Or of the surrounding pixels?
Anyway, I can't discern individual pixels when watching my tv, nor do I notice added coloring with real content. But that is, of course, just my experience and my tv/eyes/brain combination.

[iSunrise]

But what would the color be in the next frame in the same spot? Or of the surrounding pixels
I learned to go by what I know, instead of what i don´t know. It wouldn´t pose a problem though to make a shot of the next frame if it were a movie. And since the gradient shows the majority of the pixels in a greenish tone, it´s not just one single pixel, otherwise I wouldn´t have been able to see it with my eyes.

Anyway, I can't discern individual pixels when watching my tv, nor do I notice added coloring with real content. But that is, of course, just my experience and my tv/eyes/brain combination.
The dithering in madVR is currently mainly used to counter the additional processing steps madVR needs to do for it´s magic to work, to arrive at the highest possible resulting bitdepth, before it sends the RGB data to the display. Since the majority only really watches 8bit movies on an 8bit display with not too much processing, this will probably not be visible. However, if the bit-differences are getting bigger (more general processing, 3DLUTs, higher input bit-depth than output bit-depth, 6bit displays, etc.) this could eventually already be a problem.

But I don´t want to make assumptions.

It´s entirely up to madshi if he thinks this could be a problem and that´s why I posted it, to help madshi with his work. From a personal POV, I am also very much pleased with opposite_dynamic as it is, irregardless of the above anomaly I accidentally found today.

[6233638]

(2) I used GIMP to reduce the input range from 0-255 to 192-255 and the output range from 0-255 to 0-128 and applied the change
(3) In the next step I reduced the input color range from 0-255 to 0-32 and applied the changeThis is going to significantly exaggerate any color tones in the image.
You also need to average the image before you boost the contrast, rather than pick a random pixel.

The values used in 8-bit dithering should all be plus or minus one of the target value, and average out to a neutral tone.

In 8-bit, plus or minus one values should be very difficult to distinguish from each other when they are pixel-sized, randomized, and part of a moving image.


On the subject of noise, it's true that the colored builds are using more pixels for dithering, but this has the effect of creating a more continuous tone with a far less obvious texture, as there is less difference between each pixel, which means that the perceived image should have less noise.

If I disable BFI I end up with 360p motion(from what many reviews are claiming at least, and indeed it's motion-blur galore to my eyes), pretty nasty flickering and I'm stuck in 60Hz-land, I don't see how that's gonna help with anything. Especially as YV12 is 4:2:0 to begin with.But the dithering is being done in RGB, so viewing in 4:2:2 will blur it.
There's nothing wrong with using 4:2:2 for watching films, but I wouldn't use it when we are trying to judge the quality of pixel-level details.

too bad those colorful builds look blurry in motion to meI really don't see why dithering the image would introduce motion blur.

[iSunrise]

This is going to significantly exaggerate any color tones in the image.
You also need to average the image before you boost the contrast, rather than pick a random pixel.

The values used in 8-bit dithering should all be plus or minus one of the target value, and average out to a neutral tone.

In 8-bit, plus or minus one values should be very difficult to distinguish from each other when they are pixel-sized, randomized, and part of a moving image.


On the subject of noise, it's true that the colored builds are using more pixels for dithering, but this has the effect of creating a more continuous tone with a far less obvious texture, as there is less difference between each pixel, which means that the perceived image should have less noise.
It´s true of course, if I exaggerate my values, I will obviously get exaggerated results, but even without using GIMP (meaning no range/contrast adjustment at all), I can pick greenish pixels in the 1:1 shots that I did directly with madVR. Why only green? Do you have an explanation for that?

Because the green is adding up to what I presented, a greenish gray gradient. I just made it a lot more visible by reducing the range, therefore they are not as close together anymore, they are more spread out, but the greenish pixels are even in the untouched screenshots.

And while I am certainly not able to spot every greenish toned pixel in the untouched screenshots from the 16bit source I used, I am still able to see with the color picker, that (especially in the darkest tones) RGB is not 16, 16, 16 anymore, but 16,17,16 instead when compared to the opposite_dynamic build. I´m not sure why, but every pixel that differs from mono_dynamic has a more greenish tone (+1) to it. It also has a higher brightness and saturation, probably because of the additive nature of the RGB dithering. I would have at least expected redish or blueish tones, too. They are random, after all.

Other than that, I like the more pixels that are used for dithering. It seems they are worth the results we´re getting, even when I had some doubts at first.

[webs0r]

FWIW my vote goes to opposite color static.
I like how I perceive the least amount of noise, even temporally, and I kind of dig the 'dirty screen' effect - maybe because it makes me feel that I'm looking at frames closer to how they are coming out of the decoder.

Tbh though without the 4 bit mode PLUS going up close to the TV, I can't tell the difference between any of these sitting ~3.5m away from a 70" screen...

[6233638]

Because the green is adding up to what I presented, a greenish gray gradient. I just made it a lot more visible by reducing the range, therefore they are not as close together anymore, they are more spread out, but the greenish pixels are even in the untouched screenshots.In my tests, when I average out the pixels, they all end up at neutral RGB values.

As I understand it, the monochrome builds should only move up or down one "full" pixel value.

E.g.

16,16,16
17,17,17

The color builds should be blending:

16,16,16
17,16,16
16,17,16
16,16,17
16,17,17
17,16,17
17,17,16
17,17,17


This should all average out to the same tone, but with the color builds, there will be far fewer "gaps" in the dither pattern, reducing the texture and creating a more continuous tone.

[GREG1292]

I've never understood the love for projectors. Sure it's a relatively cheap way to get huge pictures but it's impossible to get as good black levels as most LCDs, let alone plasmas, even in dark rooms. I remember watching clips of LotR at a friend's house a few years ago on his projector and I wasn't impressed at all.

With the new screen technology the black levels are quite impressive and find a lot of flat panels crushed in comparison. I will take the projector over my 65" Panasonic
plasma any day which I own also. Without madVR new OPC that would be debatable. Dlp all day long as Dila and LCD are not viewable for me.
Now if we are talking white screen technology I would prefer a flat panel over any projector. I do have a batcave for my theater viewing.

[6233638]

It´s true of course, if I exaggerate my values, I will obviously get exaggerated results, but even without using GIMP (meaning no range/contrast adjustment at all), I can pick greenish pixels in the 1:1 shots that I did directly with madVR. Why only green? Do you have an explanation for that?On neutral patterns, green and magenta seem to be prioritized in the opposite build, rather than RGBCMY as seen in the colored builds.

I think it is more appropriate to desaturate the image if you are going to boost the contrast significantly to analyze the dither structure.

Here's what a value near black looks like with the contrast boosted considerably, and then desaturated:

http://abload.de/img/dither-patternst5uew.gif

Both the monochrome and opposite builds average out to a value of 143, but the colored build averages out to 146. This may simply be due to the area I selected rather than an error to be concerned about though.

Because the monochrome build only uses two shades for dithering, the texture is much more visible.

Doing this is a clear illustration of what I meant about the "opposite" build adding the least amount of texture, and creating a more continuous tone.

I kind of dig the 'dirty screen' effectDithering should not be imparting obvious artifacts on the image like this though.

Tbh though without the 4 bit mode PLUS going up close to the TV, I can't tell the difference between any of these sitting ~3.5m away from a 70" screen...That's why I think it's actually not worth getting too concerned about some of the artifacts that can be seen in the 4-bit mode, compared to the 8-bit mode, or when boosting the image contrast. Things are getting blown out of proportion.

And there are some artifacts which are obvious in the 4-bit mode, but will be significantly reduced when moving to 8-bit (e.g. the colors introduced with the colored/opposite builds) when there are some artifacts that, if visible, will not be reduced at all moving to 8-bit. ("dirty screen" effect)

[James Freeman]

Few basic question I wanted to ask for some time now, that I can't find the answers all over the web:

Queue of what? Frames?
Queue sizes, how they work together and how GPU/GPU memory is related.
What is the function of each queue (decoder, upload, render, backbuffer)?
What is the "order" of their operation?

What does the "Use Separate Device (Vista & Newer)" do?
Is it of any benefit to performance?

I think this information will be valuable to many others.
Thank you.

[ryrynz]

I can't find the answers all over the web:

What does the "Use Separate Device (Vista & Newer)" do?
Is it of any benefit to performance?


Maybe all over the web didn't include this forum :P In my own testing on Intel hardware it does improve performance a little.

Here's a special new test build which *may* help reducing presentation glitches with high refresh rates:

This test build contains a wild hack which only works properly if you set "after last render step" to "flush & wait". Without that you'll get nasty results.


madVR v0.59 released
* renamed glitch hack to "use a separate device for presentation"

[James Freeman]

Thank you ryrynz.

How about queues then?