madVR - high quality video renderer (GPU assisted)

[seiyafan]

Can someone provide a picture comparison between ordered dithering and ED? I can't seem to notice any difference but maybe I wasn't looking hard enough.

[flashmozzg]

Originally Posted by seiyafan

Can someone provide a picture comparison between ordered dithering and ED? I can't seem to notice any difference but maybe I wasn't looking hard enough.

In 8+ bit content the difference is very small and even less obvious to the eye. The only way to see it clearly is to use special patterns and high contrast.

[Andy o]

Originally Posted by 6233638

All the Kuros support 0-255.


Edit: Actually, I'm not 100% certain if the US consumer models do - they were stripped down compared to the ones sold elsewhere, to sell people on the "Elite" models.
Outside USA there was only one range of consumer sets, equivalent to the "Elite" without the branding, if I remember correctly.

The KRP-600M, being the pro monitor, absolutely does though.

I got the 500M, and it does indeed. AFAIK the "A" models sold elsewhere were proper TVs though with tuners, though apparently the speakers were still optional.

[tjcinnamon]

Is there a method where MadVR can determine what type of video it is and select the settings based off of that?

Example: If I am running 1080p video then I know I won't use Luma Image Doubling and therefore can put more GPU processing power into the Chroma Upscaling. However, if my source video has lower resolution I can reduce the Chroma Upscaling because I know I will be putting more processing power into Image Doubling.

Hope that exists and my description makes sense.

[nevcairiel]

madVR has profiles just for that. Check the first couple posts in this thread, madshi has a small guide there how to use them.

[tjcinnamon]

Originally Posted by tjcinnamon

Sweet! I created the profile groups no problem. I read the post a few times but can't figure out where to put the script.

Any ideas?

never mind

http://i.imgur.com/7lFvOsn.jpg

[tjcinnamon]

I'm having issues scouring the forums trying to find out the difference between Chroma Upscaling, Image Upscaling, and Image doubling.

If I have a 1080p video with a 1080p display I can leave Image Doubling off correct?

If I have a 720p video with a 1080p display, should I set my Image doubling to 2.0, 1.5, 1.2, or Always? How many neurons (32?)

What will give me the most "bang for my (GPU) buck"? Image Upscaling or Chroma Uplscaling? Do I need Image upscaling with 1080p? I'm trying to set up profiles.

Thanks,
JOe K.

[pie1394]

Originally Posted by tjcinnamon

I'm having issues scouring the forums trying to find out the difference between Chroma Upscaling, Image Upscaling, and Image doubling.

If I have a 1080p video with a 1080p display I can leave Image Doubling off correct?

It does not matter since no such scaling is actually performed...

Originally Posted by tjcinnamon

If I have a 720p video with a 1080p display, should I set my Image doubling to 2.0, 1.5, 1.2, or Always? How many neurons (32?)

1280 (source) * 1.5 = 1920 (target) ... so the answer should have been very clear.

64 is somewhat better if the GPU power is affordable w/o crazy noise. Otherwise 32 is indeed recommended for Luma. I also set the Chroma Upscaling with NNEDI3_16 regardless of contents. For 720 content to 1080, Image Upscaling is set the Lanczos4AR, and Image Down-scaling is set to Spline3AR_LL -- according to some discussion about the negative text border effect of Lanczos3LL or Lanczos3AR.

Originally Posted by tjcinnamon

What will give me the most "bang for my (GPU) buck"? Image Upscaling or Chroma Uplscaling? Do I need Image upscaling with 1080p? I'm trying to set up profiles.

Thanks,
JOe K.

In my memory, the HD7970 GPU's upper limit is the Debanding_angle_detection + NoDithering + Chroma Upscaling NNEDI3_128 for 1080p24 content to 1080p24 output signal.

If you want to run 1080p24 content's Chroma Upscaling with NNEDI3_256, I guess R9 290X or GTX Titan might be required...

ps: Don't forget the workload of 1080p resolution is exponential times of 720p, and again for 480p / 360p . For the 50i / 60i interlaced contents, it makes the GPU work-load about 2.2 ~ 2.5x times of 24fps film contents with vector-based-pixel-adaptive + frame-doubling deinterlacing.

[Anime Viewer]

Originally Posted by tjcinnamon

I'm having issues scouring the forums trying to find out the difference between Chroma Upscaling, Image Upscaling, and Image doubling.

If I have a 1080p video with a 1080p display I can leave Image Doubling off correct?

If I have a 720p video with a 1080p display, should I set my Image doubling to 2.0, 1.5, 1.2, or Always? How many neurons (32?)

What will give me the most "bang for my (GPU) buck"? Image Upscaling or Chroma Uplscaling? Do I need Image upscaling with 1080p? I'm trying to set up profiles.

Thanks,
JOe K.

Contrary to what some people have told you Image Doubling can still be active when playing a 1080p video on a 1080p monitor (even thought it technically shouldn't be). Your best off having a 1080 profile and disabling it there like you mentioned.
With the resources you free up disabling Image doubling you could enable Error diffusion or up neurons in upscaling (if you choose to use NNEDI3 for that).

Most people have said they don't see significant (if any) benefit from using image doubling with hd content going from 720->1080. My suggestion is experiment by playing part of a 720p video with it enabled, and then play that same part with it disabled. If you notice a difference that is beneficial with it enabled then use it, but if you do not then disable it to free up those resources.

Unlike Image Doubling which can be unchecked there is no way to disable either of the upscaling configurations in madVR. Most people here would probably recommend setting it to Jinc 3 taps if your system can handle it, and keep low render numbers. It depends on your graphics card/chip, but most people (with decent gpu(s)) will see the lowest render times with good quality selecting DXVA2 for image upscaling. My recommendation would be to make a profile for 1080 Image Upscaling, and try DXVA2. If you see pixelation or a drop in perceived image quality then you could switch back to Jinc, but DXVA2 would be the closest you'll get to disabling it.

[Meulen92]

From first hand experience I can say I can see a significant improvement in image doubled 720p content over upscaled 720p content when watching on a 1080p display. I use 128 neurons. For example hair on peoples arms is rendered much sharper in my opinion.
My settings are: chroma upscaling: nnedi3 128 neurons, image downscaling: cr ar ll, image upscaling: jinc3ar, dithering: ED2 + both options.

[tjcinnamon]

WOW! Thanks for the in-depth responses everyone!

So Image-Doubling simply increases the resolution. If that's the case, what does Image-Upscaling do? Is that more for brightness (because chroma handles color)?

[huhn]

Chroma resizing in MadVr is only use to turn not 4:4:4 YCbCr into 4:4:4 YCbCr nothing else.
image upscaling/downscaling is used for image upscaling/downscaling(i'm so genius i know) it takes the image after image doubling and scales it to the target resolution.

read this: http://en.wikipedia.org/wiki/Chroma_subsampling

all BD and DVD are subsampled to 4:2:0

[Anime Viewer]

Originally Posted by Meulen92

From first hand experience I can say I can see a significant improvement in image doubled 720p content over upscaled 720p content when watching on a 1080p display. I use 128 neurons. For example hair on peoples arms is rendered much sharper in my opinion.
My settings are: chroma upscaling: nnedi3 128 neurons, image downscaling: cr ar ll, image upscaling: jinc3ar, dithering: ED2 + both options.

128 seems unnecessary to me. The linked image shows luma image doubling with 32 neurons enabled on the left side, and image doubling off on the right side. With it off the aliasing on the lips, teeth, and circles are noticeable, but I don't notice any difference between the 32 shown in the left picture and any other image double setting be it 16 or 256. Simply enabling it at 16 neurons seems like enough.
Image Doubling on vs off

On the topic of wasted resources. I did some experimenting, and it seems like a waste to use Jinc for Image Upscaling. The image on the left (below) uses bilinear image upscaling while the one on the right (below) uses jinc image upscaling. I don't see a significant difference, and going with bilinear drops the render times by around 10ms.
Bilinear vs Jinc image upscaling

[michkrol]

Originally Posted by Anime Viewer

128 seems unnecessary to me. The linked image shows luma image doubling with 32 neurons enabled on the left side, and image doubling off on the right side. With it off the aliasing on the lips, teeth, and circles are noticeable, but I don't notice any difference between the 32 shown in the left picture and any other image double setting be it 16 or 256. Simply enabling it at 16 neurons seems like enough.
Image Doubling on vs off

As always with these tests, the results can and will vary based on source material type, resolution and quality. For anime with thick contours, like in your example just enabling NNEDI works wonders, but live-action imagery can benefit from more neurons.

Originally Posted by Anime Viewer

On the topic of wasted resources. I did some experimenting, and it seems like a waste to use Jinc for Image Upscaling. The image on the left (below) uses bilinear image upscaling while the one on the right (below) uses jinc image upscaling. I don't see a significant difference, and going with bilinear drops the render times by around 10ms.
Bilinear vs Jinc image upscaling

The difference depends mostly on scaling factor, so let me ask, are you using NNEDI doubling for that comparison? If yes, then the scaling factor is ~1.11x meaning you'll only notice switching upscaling to nearest neighbor.
Sometimes, depending on the source, you can get a (subjectively) better image by using cheaper algorithms, because they blur details and hide noise that shouldn't be there in the first place.

All in all, for your example you're right, but you can't say your findings are true for every video out there

[mindbomb]

I think I may have found a bug - the "use opencl to process dxva nv12 surfaces" option is active when film deinterlacing is used, and not just for video deinterlacing.

[Asmodian]

Why is that a bug? That option should be active anytime DXVA decoding is used.

[magnusr]

Hi

I have panasonic vt50 (16-235) plasma. Windows 8 using ATI 280x with latest driver. Using mphc, lav and madvr.

Is the correct settings:
ATI CCC: pixel format: RGB 4:4:4 (Limited RGB)
ATI CCC: Video: Dymanic range not selected
LAV Video Decoder: RGB output leves: untouched
madvr: RGB output leves: TV levels (16-235)

Is this correct? Thx in advance

[huhn]

this setting is right if you only care for MadVR use.
just test it with a black clipping file.

can't those tv do 0-255.

[magnusr]

Originally Posted by huhn

this setting is right if you only care for MadVR use.
just test it with a black clipping file.

can't those tv do 0-255.

Thanks So I guess its correct for my mkvs then.

But what do you mean if it's only for madvr use? You mean bad for desktop usage?

[DarkSpace]

Originally Posted by magnusr

ATI CCC: pixel format: RGB 4:4:4 (Limited RGB)
[...]
madvr: RGB output leves: TV levels (16-235)

This looks wrong. Assuming that your TV can't handle fullrange RGB, you should set it to either

ATI CCC: pixel format: RGB 4:4:4 (Limited RGB)

madvr: RGB output leves: PC levels (0-255)

(which is the general-purpose solution), or to

ATI CCC: pixel format: RGB 4:4:4 (Full RGB)

madvr: RGB output leves: TV levels (16-235)

(which is probably the preferred solution if you care only about your videos displaying correctly - be warned: anything other than madVR will display incorrectly, with clipped colors).

In your current setup, madVR squeezes the 0-255 RGB data into the 16-235 range, and later, your video driver squeezes that 16-235 data, which it thinks is 0-255 data, into 16-235 again!