madVR - high quality video renderer (GPU assisted)

[madshi]

Ah, ok, that makes sense. Maybe it should be reported to x264 as a bug. If x264 isn't able to subsample chroma correctly then it should error out with RGB input. Producing incorrect results is a bad idea, IMHO.

[6233638]

Originally Posted by madshi

Sorry, but I can't take any website serious if they claim that Bicubic looks better than NNEDI3. That's more than ridiculous. Either they've totally screwed up their tests/measurements, or their test methods are not capable of properly measuring what looks good to our eyes, or they intentionally manipulated the test results.

This is why PSNR is not really used for this type of comparison any more. SSIM generally produces more perceptually correct results, but you still need to do a visual comparison.

Originally Posted by madshi

So many other things that need attention at the moment.

Originally Posted by madshi

castle 2x NNEDI3

Hmm...

[huhn]

Originally Posted by madshi

Ah, ok, that makes sense. Maybe it should be reported to x264 as a bug. If x264 isn't able to subsample chroma correctly then it should error out with RGB input. Producing incorrect results is a bad idea, IMHO.

but it can encode in rgb directly is bad but it is capable of it.
that's what i don't get he inputs rgb but it encodes YCbCr 4:2:0.
and what conversion is used for rgb -> yua bt601, bt709 or nothing??

like nevcairiel said avisynth and add the right tags to it.

[madshi]

Originally Posted by 6233638

Hmm...

Shushh...

[nevcairiel]

Originally Posted by huhn

that's what i don't get he inputs rgb but it encodes YCbCr 4:2:0.

It always does that, unless you actually tell it to encode in 4:4:4 or RGB directly. I assume thats done to produce the most compatible files.

[turbojet]

I haven't seen 4:2:2 telecined source either. It would be nice if madvr could work around conversion to 4:2:2 though.

While I don't interpolate myself, I do occasionally decimate with avisynth because madvr's ivtc doesn't handle some situations such as this simple one which is telecined material that was deinterlaced so every fifth frame is a duplicate. It would be nice if this would work without disabling display changer and smoothmotion. I would think MadVR would be able to handle this. I suppose I could ask in the ffdshow thread, resizing avisynth works fine with player and madvr, is this being passed from ffdshow or is it calculated by player/madvr?

On bigger screens a lot of grain can be unbearable. NCIS: LA has a lot of grain, I'll try to remember to cut a sample of it when it airs again in a few weeks. I haven't messed with any avisynth filters for it but will do, after seeing madvr's deband vs avsiynth methods, it would be interesting to see what madshi could do when/if he finds time for something like this though. One thing about madvr's deband is it doesn't add grain, at least not much, while avisynth deband filters do, but in most cases the no added grain is desirable imo. I really don't get dynamic contrast, it's become very common feature on lcd's but it looks horrible imo, lcd's need to work on getting deeper blacks, not brighter.

I don't think still pictures do nnedi3 much justice, while there are definitely some pros and a few cons of the still picture castle comparison to bicubic. At 24fps those cons pretty much disappear IME.

[James Freeman]

I have dug dipper into the chroma analysis, and....

Yes, the actual chroma on a blu-ray is shifted 1 pixel to the right.
And when selecting a Chroma Upscaler in MadVR this moves the chroma 1 pixel to the left where it should be. (yes I feel stupid right now)

I also included a nice video of my reference blu ray to show you guys how bad (low-res) a blu-ray chroma really is,
And how of a good job MadVR Chroma Upsampling really does.

Please download this clip and bask you eyes in amazement of:
The Ugliness of Blu Ray Chroma

Well.. you learn something new every day.. only I do it loudly.

[huhn]

While I don't interpolate myself, I do occasionally decimate with avisynth because madvr's ivtc doesn't handle some situations such as this simple one which is telecined material that was deinterlaced so every fifth frame is a duplicate. It would be nice if this would work without disabling display changer and smoothmotion.

that would be inverse telecine (edit: field matched is the right word) but this sample is not inverse telecine it is 3 frames plus 2 >bob< frames. madvr gets 29p and handle it like true 29p nothing to fix there just wrong/bad encoded.

[pirlouy]

Originally Posted by madshi

Dynamic Contrast Enhancement
I believe such an algorithm is bad. What could make sense is to watch a video over a longer time period to detect the overall range the video is encoded in (e.g. fullrange, limited range or even double/triple expanded) and then use that information to automatically adjust madVR's input levels. But adjusting such things *per frame* will in many cases lead to image pumping and other artifacts.

A lot of TV offer this. This setting allows a better visual. I guess it have some drawbacks, but no artifacts. Sure it does not respect the source, but it's a feeling: image looks better.

On LCD, the other dynamic contrast called "dynamic contrast" is a problem, and at least on my TV, he can't be disabled. I suspect it of causing flickering.

[Razoola]

I don't like the idea of these enhancements (IFC, DCE etc). My panel has many options like this and I have them all turned off.

I must admit the idea of an upscaler that takes a few frames of information into account when up scaling sounds interesting. I also would also be interested in experimentation of forcing a panel that supports 3D at 120hz to display a normal 2D picture at that rate. Its just not possible to put a 3D panel into a 120hz refresh rate via windows unless its a dedicated PC monitor. I think most only go to 120hz when they have 3d material sent down the HDMI. So what Im suggesting there is to send a standard 2D 24fps frame to the panel as 3D (make the frame the same for both the left and right eye). I think that would be intresting because plasmas at least do not have options to watch 2D at 120hz unless you activate IFC but that sucks. I guess it may allow for picture improvement (esp in motion scenes) but then again it might be worse.

[turbojet]

Originally Posted by huhn

that would be inverse telecine (edit: field matched is the right word) but this sample is not inverse telecine it is 3 frames plus 2 >bob< frames. madvr gets 29p and handle it like true 29p nothing to fix there just wrong/bad encoded.

It is badly encoded and a bit much to expect madvr to find and remove dupes by default but using deint=ivtc I'd expect to handle it, the pattern doesn't get any simpler. Instead madvr does an unnecessary deinterlace and chooses the wrong frames to remove.

[magus]

Originally Posted by Razoola

I also would also be interested in experimentation of forcing a panel that supports 3D at 120hz to display a normal 2D picture at that rate. Its just not possible to put a 3D panel into a 120hz refresh rate via windows unless its a dedicated PC monitor. I think most only go to 120hz when they have 3d material sent down the HDMI.

I don't know if this has anything to do with what you said but I have a 120hz monitor and when I watch movies on it madvr says the display is 120hz and composition rate is also 120hz(on the cntrl+j screen). Also does anyone know if there is an advantage in watching movies in 120hz. Right now I just use it for games which is nice.

Originally Posted by madshi

Not sure why. But if the NVidia stereoscopic 3d mode makes problems, that's probably out of madVR's control.

If it claims you're running out of memory, you could try lowering the GPU queue size and/or the number of prepresented frames to save GPU memory (if the complaint is about GPU memory, otherwise try lowering the CPU queue size) to see if that helps.

Reducing the GPU and CPU queue did not work but turning off present frames in advance made the out of memory error go away. But turning off the present frames in advance also makes my 3d icon and 3d glasses turn on in fullscreen exclusive mode(weird, I guess nvidia thinks exclusive mode with present frames in advance turned off is a 3d application?). I also noticed that the average stats for rendering and present are higher when 3d mode is forced on(by turning off present frames in advance) even though it doesn't say out of memory anymore.

Anyway I figured the best option as of now is to turn off all 3d related options in nvidia when I'm not playing games.

[leeperry]

Originally Posted by Razoola

I must admit the idea of an upscaler that takes a few frames of information into account when up scaling sounds interesting.

vReveal is the only consumer software that enhances video with genuine super-resolution technology. It uses patented algorithms to analyze and extract information from multiple frames to reconstruct single, enhanced frame. vReveal has the unique ability to increase resolution and remove noise in all your videos.

This thing can run in realtime apparently, too bad vReveal might very well have gone belly up as their domain name is a 404 now. This woulda killed in mVR

[huhn]

Originally Posted by turbojet

It is badly encoded and a bit much to expect madvr to find and remove dupes by default but using deint=ivtc I'd expect to handle it, the pattern doesn't get any simpler. Instead madvr does an unnecessary deinterlace and chooses the wrong frames to remove.

that's not the point of ivtc the sample is progessive and nothing else.

ivtc is not simply removing duplicates if it does this terrible thinks will hppen anime is from time to time 5 fps or even lower that'S not how ivtc works.

ivtc removes duplicates from filedmatching but how to field match a progressive source? use it on interlaced videos and nothing else.

[Guest]

Originally Posted by huhn

use it on interlaced videos and nothing else.

More precisely, use it on telecined video.

[Xaurus]

Originally Posted by Razoola

I don't like the idea of these enhancements (IFC, DCE etc). My panel has many options like this and I have them all turned off.

I agree. Hopefully madshi won't waste time on stuff like that.

[Razoola]

Originally Posted by magus

I don't know if this has anything to do with what you said but I have a 120hz monitor and when I watch movies on it madvr says the display is 120hz and composition rate is also 120hz(on the cntrl+j screen). Also does anyone know if there is an advantage in watching movies in 120hz. Right now I just use it for games which is nice.

I also have a 120hz pc dedicated monitor. If you are using 2 displays though with AERO enabled you cannot take the composition rate as the refresh rate on the 2nd display. The composition rate is always the refresh rate of the primary display. My 2nd display is a 3D plasma and although madVR reports 120hz composition rate the actual refresh rate of the panel is 24hz (as reported by the panel, windows and madVR in the display value).

[yok833]

Originally Posted by Xaurus

I agree. Hopefully madshi won't waste time on stuff like that.

It depends of how would be the result with new algorithms from madshi... It is turned off on your TV because you are not happy with the results... Me, I would be happy of each new option that could improve the picture quality of my movies

[Razoola]

We have those turned off because we want our picture to match how the director of the movie wanted them to be. Once you start changing settings adding dynamic contrast etc you begin to ruin the experience of getting lost in the film experience. This is because the atmosphere the director is trying to create with lighting is lost.

[yok833]

Originally Posted by Razoola

We have those turned off because we want our picture to match how the director of the movie wanted them to be. Once you start changing settings adding dynamic contrast etc you begin to ruin the experience of getting lost in the film experience. This is because the atmosphere the director is trying to create with lighting is lost.

I totally agree with you about respecting the atmosphere of the director.. but is it incompatible ? Depending of your TV or your source, a noise reduction, a sharpen filter or a dynamic contrast algorithm (specially cooked by madshi!!)... can maybe help in getting closer of the original atmosphere?