madVR - high quality video renderer (GPU assisted)

[NicolasRobidoux]

The most natural looking downsampler I know, for DSLR image downsampling, is EWA quadratic B-spline-windowed Jinc 3 (with no deblur, although the same deblur as madVR's Jinc 3 does not break anything), through linear light.
The first image I try when comparing downsampling methods is http://upload.wikimedia.org/wikipedia/commons/8/85/Calliphora_sp_Portrait.jpg which I downsize to 403x600.
Often, this is enough to disqualify a downsampler right there and then.
Also, anything but linear light is deadly.
Of course, making this run reasonably fast when downsampling would take a good amount of "craft".
And I have not checked how well it does with things like text. This being said, this is a scheme with fairly low ringing (without AR).
P.S. Examples of enlargements with this scheme, unfortunately with sigmoidization, which possibly only works well when the source material can "nicely" be converted to linear light without gamut "mismatch", are here: http://www.imagemagick.org/discourse-server/viewtopic.php?f=22&t=21933.
P.S.2 Downsampling examples are here: http://www.imagemagick.org/discourse-server/viewtopic.php?f=1&t=20992&start=225#p90394

[leeperry]

I would love to test it for you but a few minutes after my testing the SSD with Windows on it in my HTPC died. [..] you can discard my numbers for now.
Ah, bummer! I indeed read many horror stories with SSD's that would completely wipe themselves out of the blue, so I'll be fine with a ramdisk for my A/V apps until these things become more reliable.

Anyway, I'm getting this golden sample 650Ti (http://www.techpowerup.com/gpudb/b538/Gainward_GTX_650_Ti_GS.html), it's not that far from the 660 (http://www.techpowerup.com/gpudb/807/.html) and it's $100 cheaper too. I'll post my tests results but I'm quite sure it'll deliver :)

[egur]

Eric:
Here is an example of EWA LanczosSharp (a.k.a. Jinc 3) used with an alternate method, namely sigmoidization, that reduces halos and that Mathias found does not work well in some circumnstances. (And I'm going to have to double check on this, because it's worrisome. I'm actually not sure at all it would work well with Y'CbCr. And, after Mathias comments, I'm not even sure it always works well with sRGB pngs.)
http://web.cs.laurentian.ca/nrobidoux/misc/madVR/glyphsEWALanczosSharp.7p5.png

Were can I get the original?
BTW, I want to publish a set of results of my scaler and I want to make sure I use the same images as Madshi\ and yourself.
Can you/Madshi post links and tell me how much to scale?

[NicolasRobidoux]

@Eric:
http://madshi.net/MiniOrg.png

[NicolasRobidoux]

@Eric:
I really like the test images and enlargement ratios discussed here (the end of the thread drifts; stop at the dragon): http://www.imagemagick.org/discourse-server/viewtopic.php?f=22&t=21804.
But Mathias or somebody else may feel that this is not representative of video content?

[6233638]

The most natural looking downsampler I know, for DSLR image downsampling, is EWA quadratic B-spline-windowed Jinc 3 (with no deblur, although the same deblur as madVR's Jinc 3 does not break anything), through linear light.
The first image I try when comparing downsampling methods is http://upload.wikimedia.org/wikipedia/commons/8/85/Calliphora_sp_Portrait.jpg which I downsize to 403x600.
Often, this is enough to disqualify a downsampler right there and then.
Also, anything but linear light is deadly.
Of course, making this run reasonably fast when downsampling would take a good amount of "craft".
And I have not checked how well it does with things like text. This being said, this is a scheme with fairly low ringing (without AR.It certainly is awkward to get a specific size when using MPC-HC. The scaling options are limited, so I can only scale to 406x605 or 398x593 using the zoom options, and resizing the window to get a 600px tall image resulted in 402x600.

I had to disable the aspect ratio correction and manually resize the window to get 403x600:

http://www.abload.de/img/4039kopd.png

There does seem to be somewhat more aliasing/moire, but I don't think you are likely to run into that problem when scaling video at all.

With all the other "sharp" scaling algorithms in madVR, there is nasty black ringing around bright objects when using linear light downscaling (even with the anti-ringing filter) at least with the images I tested (didn't try with this) which is a big part in why I settled on Catmull-Rom with the anti-ringing filter enabled.

[NicolasRobidoux]

There does seem to be somewhat more aliasing/moire, but I don't think you are likely to run into that problem when scaling video at all. ... I settled on Catmull-Rom with the anti-ringing filter enabled.This is a good looking result. Sure, EWA QuadraticJinc 3 has less moire, but again this is a rather extreme test case. Actually, if anything is not quite right with a scheme, this test image and downsampling ratio will make it jump at you. Schemes considered "reasonable" will produce absolutely awful results.
It certainly illustrates how well AR works (Good job Mathias!) because the Catmull-Rom result manages a nudge less halo than EWA QuadraticJinc 3, and it is also a nudge sharper.
I'm impressed.
P.S. I'm VERY impressed. Squeezing this good a result out of Catmull-Rom filtering... This is a better result than with the overwhelming majority of schemes I've tested. And believe me, me and others have looked at a lot of flies.
P.S. This being said, EWA QuadraticJinc clearly wins the moire contest.

[mzso]

The majority of motion blur seen on LCDs is now caused by the human vision system due to LCDs being sample-and-hold type displays rather than being impulse-driven like CRTs were. (and Plasmas somewhat are)

This is why modern LCDs with scanning backlights can have essentially zero motion blur.

OLED is similar to LCD in that it is a sample-and-hold display technology if it's not using black frame insertion.


Interpolation will still be necessary because 24p is not nearly a high enough framerate to perceive fluid motion. (native 24p judders a lot on panning)
Wholeheartedly agree on the last part. Sceptical of the preceding parts. 24p is a blur and choppyness fest no matter what you do with it. Also I have my doubts that BFI does any gooda at 48p+.

There's no reason OLED couldn't be more like LCDs with LED backlight scanning, rather than being more like CRT/Plasma.
I see no reason why it couldn't be both. Meaning black frame insertion as a feature that could be enabled.

LG are using RGBW subpixels, and Samsung appear to be using a vertical subpixel arrangement. Neither of these are suited to use as a monitor, which expects horizontally arranged RGB stripes.

http://roricon.egloos.com/m/3810249
This could only be a problem with sub pixel anti-aliasing. And I think it's trivial to update d2d and whatnot to support new arrangement. Actually I seem to recall faintly that every order/direction of RGB is supported.

[NicolasRobidoux]

Here's the other "downsampler disqualifier": The "Backpack" http://www.mediafire.com/view/?uz8dphpj6q0y6n5#.
I normally downsample it to 1200x801. The result with QuadraticJinc 3 (through linear light, always) is here: http://www.imagemagick.org/discourse-server/viewtopic.php?f=1&t=20992&start=225#p90394

[6233638]

Here's the other "downsampler disqualifier": The "Backpack" http://www.mediafire.com/view/?uz8dphpj6q0y6n5#.
I normally downsample it to 1200x801. The result with QuadraticJinc 3 (through linear light, always) is here: http://www.imagemagick.org/discourse-server/viewtopic.php?f=1&t=20992&start=225#p90394
http://www.abload.de/img/1200k5st5.png

I keep forgetting to disable my calibration settings and dithering when taking these, meaning I have to do it several times. :o
Again, I don't think you are going to see high-frequency detail like that, even with most Blu-ray discs, so the increased aliasing isn't too much of a concern for video.


Wholeheartedly agree on the last part. Sceptical of the preceding parts. 24p is a blur and choppyness fest no matter what you do with it. Also I have my doubts that BFI does any gooda at 48p+.24p shouldn't have any motion blur other than what is in the source.

I play games at 60p, and there's still a huge difference with backlight scanning on/off with my LCD. (not a response time issue - high speed photographs don't show any motion blur on the panel)
It's an issue of perception, and the same reason why when you use a strobe light, you are able to see things clearly that would otherwise have been a blur - it's the same thing happening here with LCD/OLED. (that doesn't use BFI/Backlight Scanning)

This could only be a problem with sub pixel anti-aliasing. And I think it's trivial to update d2d and whatnot to support new arrangement. Actually I seem to recall faintly that every order/direction of RGB is supported.Pixel structure definitely affects more than just subpixel font rendering. Any kind of computer graphics tend to have problems with non-standard layouts. There are displays out there which modulate pixel brightness by only adjusting half a subpixel (or even less) and they show obvious artefacts when used as a computer display.

[Asmodian]

Anyway, I'm getting this golden sample 650Ti (http://www.techpowerup.com/gpudb/b538/Gainward_GTX_650_Ti_GS.html), it's not that far from the 660 (http://www.techpowerup.com/gpudb/807/.html) and it's $100 cheaper too

Actually the OEM 660 is not the same as the retail 660 (http://www.techpowerup.com/gpudb/895/NVIDIA_GeForce_GTX_660.html) neither of which are very close to the 660 Ti (http://www.techpowerup.com/gpudb/364/NVIDIA_GeForce_GTX_660_Ti.html). The OEM 660 has 50% more shader units (but that 650 is only ~23% slower due to clock speed) and 58% more memory bandwidth than that 650 so they aren't that close! That 650 Ti only has 63% of the texture fill and 61% of the memory bandwidth compared to the 660 Ti.

That 650 Ti should still be great for MadVR but the 660 Ti (http://www.techpowerup.com/gpudb/364/NVIDIA_GeForce_GTX_660_Ti.html) is much faster. :)

[NicolasRobidoux]

@6233638:
Thank you!
Again, this shows that Catmull-Rom + AR downsampling (through linear light: it has to) is really high quality. If it was not, it would be obvious. As it turns out, except for the better anti-aliasing of EWA QuadraticJinc, and the slightly better anti-haloing of Catmull-Rom AR, the two methods, amazingly, are quite close. Catmull-Rom + AR is also slightly sharper.
(Note: EWA QuadraticJinc is the "natural" downsampler of my arsenal. It's not as sharp as EWA LanczosSharp or EWA Lanczos Radius 3, which is almost as sharp as tensor Lanczos 3 (which is too sharp IMHO anyway, although lots of people like sharpness). It also has less halo.)
P.S. Moire drives DSLR folk nuts. So, I paid a lot of attention to it.
P.S.2 If you enlarge both, you can see the usual difference between a high quality tensor method and a high quality EWA method: The checkerboard mode is suppressed with EWA, and not with Catmull-Rom AR. Same with the fly.

[NicolasRobidoux]

I'm also quite impressed at how EWA quadratic B-spline windowed Jinc 3 (which is a linear method) manages to compete with a nonlinear method with explicit halo suppression without being much softer.

[jmone]

I've got the GTX 660 installed with 306.97 drivers and it is working fine but I want to check if I've got the correct settings esp colour space:
Setup: HTPC --> HDMI --> AVR --> HDMI --> TV/PJ (note: the TV was prof calibrated a few years ago against the Cable Box that is on the same HDMI Input to the TV from the AVR so I'm guessing it is Limited Range)

Current Colour Space on the Dedicated HTPC (eg no desktop or game useage): Limited Range in Nvidia and Full Range in madVR - When using the AVSHD 709 test patters for Black, APL and White Clipping I see flashing just above 16 (Reference Black) and just below 235 (Reference White) which I think is "Correct". I also tried using the madNvLevelsTweaker to push full range in the driver and while this worked I did not notice a change in the test pattern behaviour. When I change madVR to limited range then I see all levels of the AVSHD 709 test pattern flash regardless of if the driver is in full or limited range mode (note: I've not tested with the PJ yet that is on the second output from the AVR). My real world test BD for this is the Spinning Cube on the opening scene in Transformers as there is alot of detail on the surface of the cube that can get hidden in the "shadows" as it spins and I the extra detail is certainly noticable when madVR is at Limited.

- Any suggestions on the Colour Space in this setup?
- Any other tweaks in the nvidia control panel that should also be changed from default.

Thanks!

[FreeFall]

madshi,

I have come across a problem with Zoom Player when using MadVR.

Playback won't start on the first try when using the delay playback start until renderer queue is full option and Zoom Players automatic subtitle stream selection. It will start normally if you press play a second time or don't use the above option.

Subtitle Settings.
http://www.mediafire.com/view/?zww21g5ua3ebe1f

This only happens with video files that use PGS (Blu-ray) subtitles.

Test Video.
http://www.mediafire.com/?7ge5zvk6k99g22y

Zoom Player 8.5 Max
MadVR 0.84.3
LAV Splitter / Video (CUVID) / Audio 0.52
xy-VSFilter 3.0.0.65
GeForce GTX 580 (306.97)
Windows 7 Pro


Thanks.

FreeFall

[leeperry]

Actually the OEM 660 is not the same as the retail 660 (http://www.techpowerup.com/gpudb/895/NVIDIA_GeForce_GTX_660.html) neither of which are very close to the 660 Ti (http://www.techpowerup.com/gpudb/364/NVIDIA_GeForce_GTX_660_Ti.html). The OEM 660 has 50% more shader units (but that 650 is only ~23% slower due to clock speed) and 58% more memory bandwidth than that 650 so they aren't that close! That 650 Ti only has 63% of the texture fill and 61% of the memory bandwidth compared to the 660 Ti.

That 650 Ti should still be great for MadVR but the 660 Ti (http://www.techpowerup.com/gpudb/364/NVIDIA_GeForce_GTX_660_Ti.html) is much faster. :)
Well, I made the assumption that mVR needs crunching power more than anything else and that memory bandwidth would only be useful for FSAA and so?

Several reviews claim that the 660Ti doesn't make much sense over the vanilla 660 due to the price difference, and as far as "Floating-point performance" is concerned the GS 650Ti reaches 1545 GFlops, the retail 660 1881 GFlops and the 660Ti 2459 Gflops.

Using CUVID, my 264 GFlops(:eek:) 8800GS (http://www.techpowerup.com/gpudb/210/.html) can do full Jinc3 AR 25fps SD to 1080p(60% GPU load) and native 1080p Jinc3 AR chroma(77% GPU load), I'm hoping to be able to process 59.94fps SD, 29.97 1440x1080(with a 16/9 DAR) and 25fps/50fps 720p :)

Street prices with shipping go as follow around here:
650Ti: 130€
GS 650Ti: 140€
660: 205€
660Ti: 270€

Sharp make good LCD panels, but bad TVs. (though I would avoid anything using a Quattron UV2A panel rather than an RGB UV2A panel)
Yeah, their Quattron's measured poorly in those tests (http://translate.google.com/translate?sl=fr&tl=en&js=n&prev=_t&hl=en&ie=UTF-8&layout=2&eotf=1&u=http%3A%2F%2Fwww.lesnumeriques.com%2Ftv-televiseur%2Fsharp-lc-70le835e-p12780%2Ftest.html)(2260:1 native CR instead of >4K:1 for the regular UV²A's) and as much as their 630 serie supports 24Hz, the 730 does not and several of them are just pimped Philips OEM (http://translate.google.com/translate?sl=fr&tl=en&js=n&prev=_t&hl=en&ie=UTF-8&layout=2&eotf=1&u=http%3A%2F%2Fwww.lesnumeriques.com%2Ftv-televiseur%2Fsharp-aquos-lc-40le630e-p12086%2Ftest.html)(I guess that'd explain why their golden samples use Sharp panels ^^) from what lesnumeriques.com are saying.

They also posted close-up pictures of many LCD panels (http://translate.google.com/translate?sl=fr&tl=en&js=n&prev=_t&hl=en&ie=UTF-8&layout=2&eotf=1&u=http%3A%2F%2Fwww.lesnumeriques.com%2Ftv-televiseur%2Fdalles-decryptees-par-sous-pixels-a1547.html). Apparently, all you need is a digital camera with a macro mode...I might try that on a rainy sunday.

[leeperry]

Ok so apparently the last couple of madVR versions don't acknowledge the ICC profile set by Windows...it used to consider it in previous versions of madVR though. I think it has to do with madVR's new behavior where it resets Direct3D or something like that because the screen flickers when I resize videos into fullscreen exclusive mode and before the last couple of versions madVR didn't do that.
LUT based ICC? I just tried to burn the gamma to death via the graphic card's CLUT in the nvidia drivers (http://www.imagebam.com/image/3c14b0217061069), and mVR doesn't bypass it, whatever in windowed or FSE mode(XPSP3/8800GS/old rendering path). Non-LUT ICC's need "color managed" compliant apps indeed, but as far as I can see Argyll's CLUT's do get through.

PS: even checking "disable GPU gamma ramps" doesn't bypass the graphic card's CLUT on my system.

[6233638]

So Windows 8 is officially out now today. Is anyone else experiencing issues with madVR's refresh rate changer?

1080p24 is outputting 1080p23, and 1080p60 is outputting 1080p59. I am able to select and use 1080p24/60 on the desktop without any problems outside of madVR.

[Neeto]

Yes, I am definately using the default Windows 7 DVD Navigator. I´m not using MPC-HC though. In PotPlayer under Preferences -> Hardware -> DVD/Blu-ray in the Navigator selection I have selected "Microsoft DVD navigation (recommended)", which is also the default.

Here´s my filter chain (as configured in PotPlayer):

DVD Navigator
LAV Video Decoder
Madshi Video Renderer
LAV Audio Decoder
DirectSound Audio Renderer

(I also have LAV splitter source configured for MPEG2 PS and MPEG 2 TS.)


The strange thing is that I have basically the same filter chain as 6233638. Even the video decoder is the same (LAV). The only things that are different are the audio renderer, the graphics card drivers and the media player.


I know and that´s also how it should be. What I´m wondering though (just out of interest) is if it would be possible to use that algorithm in such a way that you could upscale e.x. with Jinc and let the anti-ringing filter remove ringing more aggressively, so that we don´t have to use SoftCubic which basically softens the whole image (if used on luma). Because that´s one of the reasons why you (or even me in some cases) use SoftCubic in the first place, to hide source artefacts (which alters the image to some extend).

I have three HTPCs - none will play DVDs with madVR & MPC-HC, but have a HP laptop with a HP version of Cyberlink loaded and it will play DVD with madVR.
If I load Cyberlink on the HTPCs they still won't play DVDs with madVR.
I've tried latest lav decoders 0.52 still no good.
This is doing my head in :mad:
I've triad the GPL Mpeg2 decoder - no better.
I really wish we could get to the bottom of this!

[Budtz]

In the madvr wiki i found this:

For NVidia GPU owners I recommend to create custom resolutions. That makes sure that the GPU output is untouched. With the default HD/SD output modes, the GPU driver stretches the data behind madVR's back which can result in banding problems.

I just added a 50hz resolution from the list. is this not enough? can banding problems still occur? what excactly is my gpu doing?

I also found this:

For NVidia custom resolutions and ATI PowerStrip tweaking, here are the official CEA timings you should use for the various 1080p formats:

Code:

Refresh Rate, Horizontal Active Pixels, Front Porch, Sync Width, Back Porch, Total Pixels, Vertical Active Pixels, Front Porch, Sync Width, Back Porch, Total Pixels, Pixel Clock
23.976Hz, 1920, 638, 44, 148, 2750, 1080, 4, 5, 36, 1125, 74.2500 / 1.001 (= 74.175824175824175824175824175824)
24.000Hz, 1920, 638, 44, 148, 2750, 1080, 4, 5, 36, 1125, 74.2500
25.000Hz, 1920, 528, 44, 148, 2640, 1080, 4, 5, 36, 1125, 74.2500
50.000Hz, 1920, 528, 44, 148, 2640, 1080, 4, 5, 36, 1125, 148.5000
59.940Hz, 1920, 88, 44, 148, 2200, 1080, 4, 5, 36, 1125, 148.5000 / 1.001 (= 148.35164835164835164835164835165)
60.000Hz, 1920, 88, 44, 148, 2200, 1080, 4, 5, 36, 1125, 148.5000

I cant seem to change all theese values in the nvidia driver. The ones that i can change are set as above by default - maybe except for the 23.976 profile. Is it correct that i dont have much use for all of this this when i use reclock and a 50hz setting? (using 50hz lets me whatch ether pal content or 24hz content without having to change settings + it seems more stable for some reason + films are a tad smother without having the so-called soap opera effect.)

[leeperry]

Also, maybe someone else already reported it but whatever page I close mVR's config on, each time I close it, the next time it will reopen on the main page of my active display under /devices/...it was pretty handy for rolling purposes when it was remembering the last active page :o

[ajp_anton]

They will hopefully have industry-leading contrast performance, though OLED does not guarantee it. Sony's OLED headset only has 10,000:1 contrast for example (5,000:1 when calibrated to reference levels) which is no better than a high end plasma.Why don't they have infinite contrast (shut off black pixels)?

[nevcairiel]

Why don't they have infinite contrast (shut off black pixels)?

Because OLEDs have no backlight, everytime they show black they shut off the pixels, there isnt really another way for them =P

[DragonQ]

Well that doesn't make sense. If the pixel is shut off then that's surely as dark as possible, unless there's bleed from the other pixels or something. I can't see how a screen with a backlight could ever be darker than an OLED. From what I've seen OLED can be brighter too.

[ajp_anton]

Because OLEDs have no backlight, everytime they show black they shut off the pixels, there isnt really another way for them =PThat's exactly what I'm asking why they don't have infinite contrast.

[6233638]

I don't know the technical reasons behind it, but they are definitely not turning the pixels off in their head-mounted displays, and a number of OLED displays are incapable of turning the pixels completely off.

The specs Samsung announced (http://techon.nikkeibp.co.jp/english/NEWS_EN/20120607/221832/) for its 55" OLED panel were only 150,000:1 at a brightness of 600cd/m2, which is roughly on par with the Pioneer KRP Plasmas. (once both are calibrated to 100cd/m2 brightness)

[mindbomb]

Question:
Have you thought about adding a performance metric along with the sharpness, ringing, and aliasing, etc. to the scaling pages?

[JarrettH]

When you watch a b&w movie, is Chroma upscaling still happening?:confused:

[nx6]

It would depend on how the video was encoded I imagine. It could be encoded as RGB or monochrome video.

[Hprd]

That's exactly what I'm asking why they don't have infinite contrast.

Well, who knows I guess, I always assumed they'd pretty much all be like this one: http://flatpanelshd.com/review.php?subaction=showfull&id=1289487180 or this one: http://www.digitalversus.com/tv-television/sony-xel-1-p3246/test.html And hopefully they are ;)

[leeperry]

BTW, I see that you added filename tags for fps overriding, would you also consider doing it for gamut/decoding matrix(and possibly AR as well) please? So I could tag my pesky REC709 gamut+matrix SD files and other ppl could tag their 25fps accelerated NTSC DVDRips and SMPTE-C/EBU for the HD conspiracy believers.

Maybe something like:
PAL = EBU gamut
HTDV = REC.709 gamut + decoding matrix
BT601 = 601 matrix
BT709 = 709 matrix
and of course EBU, SMPTE-C and REC709 for gamuts.

and something like 1.33 and 1.78 for AR overriding would cut us from remuxing with a proper DAR, which can be a PITA when you have a truckload of files and can't use ffdshow anymore(I cry a million tears ^^)

:thanks:

[madshi]

At this point, I suspect that originals with haloing may not always be sigmoidization's friend.
May be true. As you said, different color spaces do different things to different types of lines. Some thin out light lines and make dark lines fatter. Others do the opposite. Depending on the content, one can look better with one test image, another one with a different test image. It could be true that light halos generally don't fit well with sigmoidization.

FWIW, I've also found other cases where sigmoidization wasn't working too well. I don't have time to double check right now, but I *think* this image also made problems with sigmoidization:

http://madshi.net/colorsrc.png

This one only really looks good with linear light upscaling, due to the extreme dithering patterns. But I don't think there's ever going to be video content looking like this, so this specific test image doesn't bother me too much.

Mathias:
You knew that Sandy Bridge has a built-in algorithm which can roughly be described as an AntiRing system?
Yes, I'm aware of that, but I haven't had the chance to test the Sandy Bridge/Ivy Bridge algorithm yet. To be honest, from what I've read, my expectation is that it might reduce/remove ringing quite nicely, but it might come with more negative side effects to my own AR algorithm. But I'm just guessing here, I might be wrong.

Now that LAV will do DVDs, I'm wondering whether the quality is the same? None (weave) in lav video (pass on to gpu) vs DXVA2 in madvr, or should they be identical? :thanks:
I'm not sure what practical effect this LAV option has, so I can't say. You'd have to ask nevcairiel about this.

what would be better for luma upscaling: Lancos 3 AR or Jinc 3 noAR?
Depends on what bothers you more: ringing or aliasing. If ringing is your main enemy, choose Lanczos3 AR. If aliasing is your biggest concern, choose Jinc3.

Ok so apparently the last couple of madVR versions don't acknowledge the ICC profile set by Windows...
I haven't changed anything, other than offering the option to disable the GPU gamma ramps (see madVR settings dialog). Do you have this option activated? FWIW, I've been told that the ICC profile is ignored by the GPU when using Overlay. Maybe you're using Overlay?

This is a good looking result. Sure, EWA QuadraticJinc 3 has less moire, but again this is a rather extreme test case. Actually, if anything is not quite right with a scheme, this test image and downsampling ratio will make it jump at you. Schemes considered "reasonable" will produce absolutely awful results.
It certainly illustrates how well AR works (Good job Mathias!) because the Catmull-Rom result manages a nudge less halo than EWA QuadraticJinc 3, and it is also a nudge sharper.
I'm impressed.
P.S. I'm VERY impressed. Squeezing this good a result out of Catmull-Rom filtering... This is a better result than with the overwhelming majority of schemes I've tested. And believe me, me and others have looked at a lot of flies.
I'm glad you like it. To be honest, I didn't spend much time on the downscaling AR algorithm, so I wasn't 100% sure how good it would work. I'm happy to hear it seems to work well.

Again, this shows that Catmull-Rom + AR downsampling (through linear light: it has to) is really high quality. If it was not, it would be obvious. As it turns out, except for the better anti-aliasing of EWA QuadraticJinc, and the slightly better anti-haloing of Catmull-Rom AR, the two methods, amazingly, are quite close. Catmull-Rom + AR is also slightly sharper.
(Note: EWA QuadraticJinc is the "natural" downsampler of my arsenal. It's not as sharp as EWA LanczosSharp or EWA Lanczos Radius 3, which is almost as sharp as tensor Lanczos 3 (which is too sharp IMHO anyway, although lots of people like sharpness). It also has less halo.)
P.S. Moire drives DSLR folk nuts. So, I paid a lot of attention to it.
P.S.2 If you enlarge both, you can see the usual difference between a high quality tensor method and a high quality EWA method: The checkerboard mode is suppressed with EWA, and not with Catmull-Rom AR. Same with the fly.
Why are you using EWA QuadraticJinc instead of EWA LanczosSharp? Is the ringing the only reason? Unfortunately madVR doesn't do "Jinc" downscaling at the moment. I'm not sure if I'm going to add it. But I guess Jinc AR for downscaling might be the ultimate downscaling solution? It would be rather slow, though (for realtime video playback).

I've got the GTX 660 installed with 306.97 drivers and it is working fine but I want to check if I've got the correct settings esp colour space:
Setup: HTPC --> HDMI --> AVR --> HDMI --> TV/PJ (note: the TV was prof calibrated a few years ago against the Cable Box that is on the same HDMI Input to the TV from the AVR so I'm guessing it is Limited Range)

Current Colour Space on the Dedicated HTPC (eg no desktop or game useage): Limited Range in Nvidia and Full Range in madVR - When using the AVSHD 709 test patters for Black, APL and White Clipping I see flashing just above 16 (Reference Black) and just below 235 (Reference White) which I think is "Correct". I also tried using the madNvLevelsTweaker to push full range in the driver and while this worked I did not notice a change in the test pattern behaviour. When I change madVR to limited range then I see all levels of the AVSHD 709 test pattern flash regardless of if the driver is in full or limited range mode
For best quality you should force NVidia to use full range and then you need to use the matching options (either full or limited) in both your display and madVR.

I have come across a problem with Zoom Player when using MadVR.

Playback won't start on the first try when using the delay playback start until renderer queue is full option and Zoom Players automatic subtitle stream selection. It will start normally if you press play a second time or don't use the above option.

Subtitle Settings.
http://www.mediafire.com/view/?zww21g5ua3ebe1f

This only happens with video files that use PGS (Blu-ray) subtitles.

Test Video.
http://www.mediafire.com/?7ge5zvk6k99g22y

Zoom Player 8.5 Max
MadVR 0.84.3
LAV Splitter / Video (CUVID) / Audio 0.52
xy-VSFilter 3.0.0.65
GeForce GTX 580 (306.97)
Windows 7 Pro
Will have a look at this later.

So Windows 8 is officially out now today. Is anyone else experiencing issues with madVR's refresh rate changer?

1080p24 is outputting 1080p23, and 1080p60 is outputting 1080p59. I am able to select and use 1080p24/60 on the desktop without any problems outside of madVR.
If you create a debug log I might be able to tell you more about what went wrong. Or maybe not.

Is it correct that i dont have much use for all of this this when i use reclock and a 50hz setting? (using 50hz lets me whatch ether pal content or 24hz content without having to change settings + it seems more stable for some reason + films are a tad smother without having the so-called soap opera effect.)
Getting your GPU output modes as correct as possible certainly won't harm. If you use Reclock, it might not make a difference in terms of smoothness. But e.g. watching 24fps movies in 50Hz means that the audio is sped up, too, resulting in the Mickey Mouse effect (audio pitch goes up). Also video will play back faster than real life, which is questionable, too. Of course it's all your decision. If you like it that way, that's ok with me.

Also, maybe someone else already reported it but whatever page I close mVR's config on, each time I close it, the next time it will reopen on the main page of my active display under /devices/...it was pretty handy for rolling purposes when it was remembering the last active page :o
It should remember the active page, it does for me. However, if there's an undefined device in your settings (look for a question mark), madVR will automatically switch to that instead of the last used page. You can fix that by properly telling madVR what device type that question mark device is.

Question:
Have you thought about adding a performance metric along with the sharpness, ringing, and aliasing, etc. to the scaling pages?
Yes.

When you watch a b&w movie, is Chroma upscaling still happening?:confused:
Yes, unless we're talking about custom encodings with weird settings. That said, the chroma information for b&w movies should usually be almost flat, so there won't be any visible difference between any of the chroma upsampling algorithms.

[madshi]

BTW, I see that you added filename tags for fps overriding, would you also consider doing it for gamut/decoding matrix(and possibly AR as well) please? So I could tag my pesky REC709 gamut+matrix SD files and other ppl could tag their 25fps accelerated NTSC DVDRips and SMPTE-C/EBU for the HD conspiracy believers.

Maybe something like:
PAL = EBU gamut
HTDV = REC.709 gamut + decoding matrix
BT601 = 601 matrix
BT709 = 709 matrix
and of course EBU, SMPTE-C and REC709 for gamuts.

and something like 1.33 and 1.78 for AR overriding would cut us from remuxing with a proper DAR, which can be a PITA when you have a truckload of files and can't use ffdshow anymore(I cry a million tears ^^)
I would consider that, but the tags would have to be very clear. E.g. "PAL" could be a part of a movie title, so it's not clear enough. Same with "EBU". "BT709" doesn't have this problem, but it could stand for the matrix or for the gamut or for the transfer function, so again it's not clear. For AR e.g. "4:3" would be great, but filenames don't support ":", so that can't be used. Ok, maybe 1.33 would do. But which ARs would have to be supported? If I think about custom encodes with stripped black bars, there could be a multitude of possible ARs which can make things quite complicated again.

[6233638]

If you create a debug log I might be able to tell you more about what went wrong. Or maybe not.http://www.filedropper.com/madvr-log


Unrelated to that, it also seems that the "path hack" to use an alternate DVD Navigator without replacing system files (I believe the tool just changed some registry keys to point to the new file) no longer works in Windows 8 so I have yet to find a way to get DVD playback working without Macrovision errors.

[madshi]

http://www.filedropper.com/madvr-log
The log claims that originally, before starting video playback, the display was 1080p60. Then, you started movie playback. The optimal display mode refresh rate would have been 23 for that movie. But the list of allowed modes was only 1080p24, 1080p50 and 1080p60, so madVR switched to 1080p24. The resulting display mode was 1080p24.

That's what the log says. If you actually got 1080p23 then that's probably a driver problem, because according to the log madVR clearly requested 1080p24, and after switching madVR double checked the resulting mode and again it was reported by the OS/GPU drivers as 1080p24.

[6233638]

The log claims that originally, before starting video playback, the display was 1080p60. Then, you started movie playback. The optimal display mode refresh rate would have been 23 for that movie. But the list of allowed modes was only 1080p24, 1080p50 and 1080p60, so madVR switched to 1080p24. The resulting display mode was 1080p24.

That's what the log says. If you actually got 1080p23 then that's probably a driver problem, because according to the log madVR clearly requested 1080p24, and after switching madVR double checked the resulting mode and again it was reported by the OS/GPU drivers as 1080p24.OK, thanks. Very strange.

If I select 1080p24 on the desktop, madVR reports that as being 24.0x when running windowed, but when it switches to FSE, it changes to 23.9x (same thing for 1080p60)

[FlashGordon]

I have a silent film on Blu-ray @1080i that seems to have an odd cadence... madvr's IVTC won't identify the cadence at all, it'll just say "unknown". There is very slight combing without any IVTC or deinterlacing applied, but with IVTC the combing seems to go away. Playback @ 60Hz seems to be smooth, but I get frame drops when I try @ 24Hz.

madshi, I've reuploaded a sample here: http://www.mediafire.com/?f8s0cfd641l59sy as per your request.

[leeperry]

It should remember the active page, it does for me. However, if there's an undefined device in your settings (look for a question mark), madVR will automatically switch to that instead of the last used page. You can fix that by properly telling madVR what device type that question mark device is.
Oh ok, thank you for the explanation. Maybe you could nag the user with a pop-up message box and tell him to define what his display is because that wasn't clear to me at all.

I would consider that, but the tags would have to be very clear. E.g. "PAL" could be a part of a movie title, so it's not clear enough. Same with "EBU". "BT709" doesn't have this problem, but it could stand for the matrix or for the gamut or for the transfer function, so again it's not clear. For AR e.g. "4:3" would be great, but filenames don't support ":", so that can't be used. Ok, maybe 1.33 would do. But which ARs would have to be supported? If I think about custom encodes with stripped black bars, there could be a multitude of possible ARs which can make things quite complicated again.
Well, a persistent problem with AR is the 720x576 AVI 25fps PAL files for instance.....I used to have a rule in ffdshow forcing 4:3 but now I'm out of luck. Maybe something like [AR=xx] for instance, et voilà ))

You could accept any AR and simply round up 1.33 to 4:3 and 1.78 to 16/9? That would save my day actually :cool:

Well, any movie that has ".PAL." in its filename will require EBU primaries anyway and same goes for ".HDTV." that will require REC.709 gamut+primaries.

You could go with [matrix=601] and [matrix=709]?

And for the gamuts, same story: [primaries=xx], accepting the SMPTE-C / EBU / REC.709 keywords?

And while we're at it, levels conversions would be extremely convenient too! Something like [source-levels=xx] accepting the two "PC" and "TV" keywords?

And the icing on the cake would be custom levels, as we all have files with funky levels(eg. went through several sloppy conversions for instance), so it'd be too awesome to be able to make a 40-200 to 0-255 conversion for instance! Something like [source-levels=xx-xx]...this would allow us to ditch ffdshow for good, and once the file is properly tagged no need to input them manually ever again...OMG, that would be too good http://forum-images.hardware.fr/images/perso/46249910.gif

PS script support will allow sharpening/mirroring/flipping/debanding/super-resolution upscaling and all the stuff you can already do in MPC and ffdshow, so together with levels conversion/gamut & primaries rolling via tags, this would become an entirely viable alternative to ffdshow and its nasty 8bit post-processing that is a bottleneck and a major disgrace to your 16bit pipeline :o

:thanks: again for making our dreams come true!

[mzso]

24p shouldn't have any motion blur other than what is in the source.

I play games at 60p, and there's still a huge difference with backlight scanning on/off with my LCD. (not a response time issue - high speed photographs don't show any motion blur on the panel)
It's an issue of perception, and the same reason why when you use a strobe light, you are able to see things clearly that would otherwise have been a blur - it's the same thing happening here with LCD/OLED. (that doesn't use BFI/Backlight Scanning)
Ok, so its still an insufficient framerate issue.

Sony's OLED headset only has 10,000:1 contrast for example (5,000:1 when calibrated to reference levels) which is no better than a high end plasma.
As far as I know you don't need more than 10000:1. At any moment the human eye can't perceive more. The fact that the eye can adapt to low light conditions is irrelevant because it takes like a half hour.

[mzso]

If I understand correctly there is no "image scaling" when I play 1920*1080 content on my 1920*1200 monitor in fullscreen mode.
Then why do I get dropped frames when I change to Jinc 3, from Lancos 8 (bot with AR)?
I don't touch anything else.

[dansrfe]

@madshi,

Ah yeah I didn't fully understand the disable GPU ramp setting. I had it enabled. If I correctly understand now, disabling the GPU ramps is mainly for the purposes of disabling the application of the ICC profile and letting madVR perform the color correction via yCMS and 3DLUTs? Thanks.

[cyberbeing]

ICC profiles do not perform gamut correction with madVR, only gamma and whitepoint correction through the gamma ramp. Assuming you have a calibrated GPU gamma ramp, unless your yCMS 3DLUT was created from Primary + Whitepoint + IRE Grayscale measurements with the gamma ramp disabled/linear, you wouldn't want to use madVR Overlay mode or check 'disable GPU gamma ramps' for FSE.

[leeperry]

If I understand correctly there is no "image scaling" when I play 1920*1080 content on my 1920*1200 monitor
Chroma is upscaled, because it's stored at half-resolution in YV12.

If I correctly understand now, disabling the GPU ramps is mainly for the purposes of disabling the application of the ICC profile and letting madVR perform the color correction via yCMS and 3DLUTs?
I rechecked yesterday by creating a funky graphic card's CLUT in RivaTuner and the "disable GPU ramps" doesn't reset the graphic card's CLUT, it really only does what it says...so I don't really see the point of this option tbh.

[Yoshi8765]

@HoP
By DXVA, I mean DXVA2. Does this make a difference?
If madVR doesn't support DXVA2, then the guide I used to set up MPC-HC makes no sense....
The guide I used is here (Image-intensive! and long!): http://wp.me/PrgSo-b1
Can anybody clear this up for me?

[NicolasRobidoux]

Why are you using EWA QuadraticJinc instead of EWA LanczosSharp? Is the ringing the only reason?Yes: EWA QuadraticJinc is my ImageMagick "high quality low ringing" (what I call "natural") linear scheme. (Although I must say that sigmoidization is quite good at suppressing ringing with sRGB images. I'm not including this nonlinear trick in this ranking.)
EWA LanczosSharp (without AR, which is a nonlinear tool, and which has no equivalent in ImageMagick) is great for downsampling and upsampling (so is EWA QuadraticJinc), but it creates too much halo for me to call it "natural looking".
That is: In a sense, EWA QuadraticJinc is my AR-like method without an AR-like limiter.

[starla]

But e.g. watching 24fps movies in 50Hz means that the audio is sped up, too, resulting in the Mickey Mouse effect (audio pitch goes up).

Actually Reclock is able to keep the pitch at correct level.

[cyberbeing]

so I don't really see the point of this option tbh.
The point is for disabling the GPU gamma ramp which ICC profiles are loaded to, assuming you want yCMS 3DLUT + madVR shaders only to perform color corrections.

Rivatuner is likely either using a different method for adjusting the GPU CLUT, or preventing the CLUT from being reset. I know that NVIDIA GPUs contain something like 3 different gamma ramp mechanisms (standard Windows/ICC CLUT, driver-based CLUT, low-level ramdac CLUT), which are all unique and apply on top of each other.

[Keiyakusha]

The guide I used is here (Image-intensive! and long!): http://wp.me/PrgSo-b1
Can anybody clear this up for me?
What exactly you want to hear? Here are some facts for you:
- madvr doesn't support native dxva
- dxva doesn't brings anything good (its not bad either, so lets call it useless waste of time).
- that guide you linked talks about different type of dxva. that one doesn't needs to be supported by renderer in any way. It is a feature of LAV video decoder.
- this guide maybe good in terms of providing results that do work, but its makes you do pretty complicated setup. and many things it suggests only needed to work around user's lack of knowledge and possible problems he can face. You supposed to read it and do exactly as it says. But don't try to understand it, because its like reading the book from end to beginning - at some point you'll figure out the story, but its not easy to follow.

[leeperry]

The point is for disabling the GPU gamma ramp which ICC profiles are loaded to, assuming you want yCMS 3DLUT + madVR shaders only to perform color corrections.

Rivatuner is likely either using a different method for adjusting the GPU CLUT, or preventing the CLUT from being reset. I know that NVIDIA GPUs contain something like 3 different gamma ramp mechanisms (standard Windows/ICC CLUT, driver-based CLUT, low-level ramdac CLUT), which are all unique and apply on top of each other.
I'm not too sure of how to check if a LUT loaded via ArgyllCMS gets through(I guess I could use pstrip, but this nasty hack gives black screens on unsupported nvidia's IME and I got better things to do than rebooting several times in a row.....or I could create a bogus .cal file), but what mVR would need is something like "dispwin -c" in ArgyllCMS that completely wipes the graphic card's CLUT(s), not just the gamma ramps...but I guess it would have to save it, erase it then restore it when the media player is closed....headaches ahead :o

[e-t172]

I'm not too sure of how to check if a LUT loaded via ArgyllCMS gets through(I guess I could use pstrip, but this nasty hack gives black screens on unsupported nvidia's IME and I got better things to do than rebooting several times in a row.....or I could create a bogus .cal file)

You can use the X-Rite LUT Tester (http://www.xrite.com/product_overview.aspx?ID=789&Action=support&SoftwareID=546) to check the current state of your hardware LUTs.

but what mVR would need is something like "dispwin -c" in ArgyllCMS that completely wipes the graphic card's CLUT(s), not just the gamma ramps...but I guess it would have to save it, erase it then restore it when the media player is closed....headaches ahead :o

CLUT = gamma ramps. They're the same thing. "Gamma ramps" is the Windows (MSDN) term for the hardware LUTs.