madVR - high quality video renderer (GPU assisted)

[Rinzler]

So even if my monitor is calibrated, MPC-HC will ignore this calibration? I need to change some madVR settings or what?

[Ver Greeneyes]

So even if my monitor is calibrated, MPC-HC will ignore this calibration? I need to change some madVR settings or what?
You should use ArgyllCMS' collink to create a 3DLUT. See e.g. the final steps of the guide in this thread (http://www.avsforum.com/t/1471169/madvr-argyllcms). Note that in ArgyllCMS, 'calibrating' refers to creating a set of three gamma curves for the individual color channels, which is usually loaded into the videoLUT of your GPU, whereas 'profiling' refers to creating an ICC profile from a set of test patch measurements. madVR can take care of applying both via a 3DLUT, though if you use madVR instead of the videoLUT for calibration, other applications will temporarily show up uncalibrated.

[Starks]

What is the bare minimum for madvr?

I have a spare machine with an Intel i945GM. Is Pixel Shader support simply not enough?

[agustin9]

I don't think the 945gm runs madvr, it lacks real pixel shaders. I think it only supports 2.0 and mostly software emulated.

[TheElix]

Nevermind, I'm a donkey. :)

[James Freeman]

I have tested the Upscaling Algorithms.

I have made few videos of several resolution patterns from:
http://www.belle-nuit.com/test-chart

I captured the screen with CamStudio and a Lossless codec.
After that I have converted the original 4:4:4 videos to other chroma modes, 4:2:2 and 4:2:0, using x264.


Here the fun begins.

For native HD content and HD monitor the Image Upscaling does nothing (obviously).
But for lower resolutions it matters very much (more later).

BUT, Chroma Upscaling is critical both for native HD and lower.
As HD chroma is only half the resolution in each direction (4:2:0) (960x540).
And visibly VERY lossy compared to the original 4:4:4 video.

After several hours of fun I came to the conclusion that Lanczon 3+AR is clear winner for Image & definitely Chroma (you read correctly NOT Jinc3).

Image Upscaling:
For image upscaling I have concluded that I'd rather have a little aliasing instead of lost detail.
Therefor Lanczos 3/4+AR wins here in every resolution especially lower ones like DVD.

Chroma Upscaling:
HD Chroma is "low res" compared to the luma, therefor the upscaling working all the time here, 480 to 1080 (actually 240 to 540).
Winner: Lanczos 3/4+AR

Why?
I have found that Jic3+AR gives me artifacts in HD and lower Chromas, whether Jinc3 (no AR) or Lanczos+AR did not.
And Anti-Ringing IS important, so Jinc without AR is a No-Go.

Don't worry, I am not all words.

Here are some Chroma Upscaling comparison images:

Lanczos3+AR:
http://i557.photobucket.com/albums/ss18/ilya-v/Other/Lanczos3AR_zps52d2f36e.png

Jinc3+AR:
http://i557.photobucket.com/albums/ss18/ilya-v/Other/Jinc3AR_zps18d453ad.png


Look at the Red/Cyan steps above and below the number 4.
You can see Jinc+AR has a "Shade" (Its not even in the middle) compared to the even looking Lanczos.
This is in HD chroma (960x540) and NO image scaling at all (1:1 pixel perfect).
Jinc3 without AR has no artifacts, BUT has strong ringing.

Lower res chromas like DVD have even more artifacts with Jinc.


Low-Res upscaling (DVD to HD):

Original resolution of the image I have created is 720x480.
The upscaled images are full size of my 1920x1200 monitor, then cropped.

Original:
* Image resolution is 720x480.
http://i557.photobucket.com/albums/ss18/ilya-v/DVDRes_zps8c83824f.png


Lanczos3+AR:
http://i557.photobucket.com/albums/ss18/ilya-v/Other/ImageL3AR_zpsbf496d24.png


Jinc3+AR:
http://i557.photobucket.com/albums/ss18/ilya-v/Other/ImageJ3AR_zps2eb881dd.png



All in all.
Lanczos 3/4 + AR is the winner for me for both Chroma & Image resizing.


EDIT:

I attached a zip file containing three short videos (mkv) that you can play and test Chroma and Resizing Algorithms.

1. DVD Resolution = for Image Upscaling comparison.
2. DVD Chroma 4:2:0 = For Chroma Upscaling comparison.
3. DVD Chroma 4:4:4 = Original Chroma for comparison.


How To:
Pause the video and go crazy with the Upscaling modes.



Edit2:
All three versions 1080p included can be downloaded from here:
http://www.mediafire.com/folder/xdwi61cstb3d3/Chroma_%2B_Resolution_-_Test_Patterns

or here:
http://www.4shared.com/folder/zsodQz1m/Chroma__Resolution_-_Test_Patt.html

[vivan]

All in all.
Lanczos 3/4 + AR is the winner for me.You forgot to add "if I'm watching only such test videos" ;)

[leeperry]

Fair enough! Well, J3AR is such a GPU hog on my HD7850 that I'll try this recommendation on live content as it can easily reach +80% load on 30fps 1080p chroma upscale + FRC + custom PS shaders(65% for 24fps) :o

BTW, PotP's coder has been kind enough to look into a few bugs when using its D3D GUI so soon enough it'll be entirely usable with mVR and that's great news as I've been looking forward to be able to use it without breaking mVR's FSE for a while, all in good time :)

[James Freeman]

You forgot to add "if I'm watching only such test videos" ;)

Your absolutely right.

I see NO difference what so ever in real videos.

But hey, the devil is in the details. :devil:
and we are using madvr and talking about chromas for a reason. ;)

[6233638]

Image Upscaling:
For image upscaling I have concluded that I'd rather have a little aliasing instead of lost detail.
Therefor Lanczos 3/4+AR wins here in every resolution especially lower ones like DVD.To each their own. This is the opposite of what I prefer. Aliasing is the first thing that takes me out of the experience. It's a harsh "digital" artifact of poorer scaling algorithms that really stands out.
When you are simply watching a film, a slightly softer image is easily ignored, aliasing is not.

Lanczos3+AR: http://i557.photobucket.com/albums/ss18/ilya-v/Other/Lanczos3AR_zps52d2f36e.png
Jinc3+AR: http://i557.photobucket.com/albums/ss18/ilya-v/Other/Jinc3AR_zps18d453ad.png

Look at the Red/Cyan steps above the number 4.
You can see Jinc+AR has a "Shade" (Its not even in the middle) compared to the even looking Lanczos.
This is in HD chroma (960x540) and NO image scaling at all (1:1 pixel perfect).
Jinc3 without AR has no artifacts, BUT has strong ringing.
This is very strange, and I wonder if it's something to do with the source. Nothing like that showed up in any of the chroma test patterns or real-world sources I used when testing chroma scaling.
But artificial patterns are not always the best thing to use when testing. If you only looked at a chroma burst pattern, you would conclude that Lanczos 8 with no anti-ringing is the best choice for chroma.
In the real world, there is very little difference - but again, Jinc 3 AR is going to guarantee there is less chance of aliasing than Lanczos 3 AR.

Overall though, when considering image quality and performance, there is basically no reason to use anything more than Bicubic 75+AR for chroma.
There are one or two rare exceptions, but those would look bad with Lanczos 3 AR or Jinc 3 AR too.

Low-Res upscaling (DVD to HD):

Original resolution of the image I have created is 720x480.
The upscaled images are full size of my 1920x1200 monitor, then cropped.

Original: http://i557.photobucket.com/albums/ss18/ilya-v/DVDRes_zps8c83824f.png * Image resolution is 720x480.
Lanczos3+AR: http://i557.photobucket.com/albums/ss18/ilya-v/Other/ImageL3AR_zpsbf496d24.png
Jinc3+AR: http://i557.photobucket.com/albums/ss18/ilya-v/Other/ImageJ3AR_zps2eb881dd.png

All in all.
Lanczos 3/4 + AR is the winner for me.Here's the issue with those tests: real-world video, especially SD video, does not contain anything nearly as high frequency as your test patterns. The pattern in the middle which shows detail loss, is rarely - if ever - going to turn up with a DVD.
The significant reduction in aliasing on the X, however, is something that has considerable real-world benefits when upscaling video.

When upscaling SD, aliasing, then ringing, are by far the biggest factors for image quality. Most SD sources have been so compressed/filtered that there is very little high frequency detail remaining, just artifacts. In many cases it is actually better for image quality to lose the highest frequency information from the image, because that is typically going to be mosquito noise.
For these reasons, I prefer to use either Jinc 3+AR, or SoftCubic 80 when upscaling SD video.

When upscaling 720p content to 1080p, Lanczos 3 AR is a reasonable choice (aliasing is generally less of a concern) though I still prefer to use Jinc 3+AR. SoftCubic is generally too soft with HD content.

To use an old example:
Soft filters are not necessarily bad for Luma. Unlike image processing where you might be upsampling a high quality photograph, the quality of commercially released DVDs is generally terrible. They have had all the high-frequency detail filtered out of them, and sharpened with very basic techniques that introduce a lot of ringing. Older discs also exhibit a lot of "mosquito" noise.

What I like about softer filters, is that they often blur out this offensive ringing/noise (note: I have no problem with film grain) while retaining most of the actual useful picture information.

This is perhaps going too far in the opposite direction, but SoftCubic 80 compared with Jinc 3 anti-ring:
http://www.abload.de/thumb/softcubici8rz3.jpg (http://www.abload.de/img/softcubici8rz3.jpg) http://www.abload.de/thumb/jinc55odk.jpg (http://www.abload.de/img/jinc55odk.jpg)

Source (http://www.abload.de/img/screenshotonpfh.png)
If you look at the two images, obviously one is a lot softer than the other - but you get over that softness very quickly when viewed on its own, and you are actually watching the content.

There is very little - if any - actual detail lost from using SoftCubic, but there is considerably less ringing and sharpening artifacts - and that's when compared to Jinc 3 AR, not Lanczos 3 AR which looks worse.

[James Freeman]

@6233638
To each their own. This is the opposite of what I prefer. Aliasing is the first thing that takes me out of the experience.

Agreed, to each his own.

As I stated, I like the sharper look even if it has visible aliasing.

This is very strange, and I wonder if it's something to do with the source.

No, the source is a capture of the original image HD with pristine 4:4:4 and 1:1 pixel resolution.

In the real world, there is very little difference - but again, Jinc 3 AR is going to guarantee there is less chance of aliasing than Lanczos 3 AR.

Again,
I care for aliasing less than what I care for sharpness and lack of artifacts.

And currently Jinc3/4+AR giving me artifacts and soft image at any chroma resolution.



In my post I have attached the sample videos so that people can judge for themselves.

I can attach a 720p and 1080p versions if anyone cares.

[James Freeman]

All three versions (480p, 720p, 1080p) are up for download in my big post:
http://forum.doom9.org/showthread.php?p=1650858#post1650858


EDIT:

After doing some more comprehensive video viewing and Upscaling comparisons.

I conclude (yet again) that:

1. No movie is ever going to reach the clarity and uniformity of the patterns in the test charts (already stated by 6233638).
2. The difference between Upscalers in real video content is so minute (even between Bilinear and Jinc),
that worrying about the smallest detail in the test pattern is ridiculous and ignorantly a time wasting hobby.

In which, the upscaler settings should be "set it and forget it" type of tweak.
If your machine is strong enough just leave it at Jinc3+AR or Lanczos3+AR and forget it.

[Rinzler]

guys, what to choose?
http://puu.sh/54WSC.png

[James Freeman]

guys, what to choose?


Using a PC monitor?
PC 0-255

PC -> TV through DVI/VGA cable?
PC 0-255

PC -> TV through an HDMI cable?
TV 16-235


Nvidia or ATI?
Use your TV as a PC monitor?
OR connect once in a while to watch some movies?

[Rinzler]

Using PC monitor mostly. Sometimes connect TV to watch movie. Don't have discrete graphics card, using CPU integrated video HD 4000.

[James Freeman]

@Rinzler

Use PC 0-255.


EDIT:
Sorry for that.

[Rinzler]

What if watching a movie from an external HDD drive connected to the TV through USB port? What happens in this case?

[dansrfe]

I still don't understand why the legacy 16-235 option still exists in driver control panels. This was originally created for CRTs, was it not?

[e-t172]

Well I could agree that 16-235 does not really have any raison d'être anymore, but it's still the standard for video (for better or for worse), so of course it should be a driver option since most projectors and HDTVs do expect 16-235.

[turbojet]

6233638: I meant while using avgDif and maxDif. For instance is there any video that looks noticeably better at 2.0/4.0/4.0 then 2.0/4.0/2.0? I've found some that look much worse but none that necessarily look better, just less detail.

Ever since re-installing windows 7 I'm having an issue switching from a 50 or 72hz display to a 60hz display. The composition rate is detected as 24 or 25 hz by madvr and there's a lot of dropped frames. The only thing that fixes it is to switch to another video renderer, which plays fine, then switching back to madvr. Things that haven't worked; opening multiple madvr sessions, closing and reopening the player, reinstalling madvr, resetting madvr defaults. This used to happen rarely on old windows 7 installation but now it happens every time. Does anyone else experience this? Any other ideas to try?

[Razoola]

The official solution is to do the following:

To set 0-255 over HDMI:
Nvidia CP -> Adjust desktop size and position -> Scaling ->
Perform Scaling On: GPU

To set 16-235 over HDMI:
Perform Scaling On: Display


When selecting GPU for scaling you'll see in the "Change Resolution" section that all UHD,HD,SD resolution are now under PC resolution and they retain 0-255 range.

When selecting Display for scaling you'll see in the "Change Resolution" section that all UHD,HD,SD resolution are now separated and they retain 16-235 range.



Are you 100% sure about this? On my system I do not have an option for GPU in that box for my plasma display (driver version 331.65). I only have that choice on the primary monitor display. Also, how does that affect custom resolutions.

I set to 0-255 by altering the INF before install. Does this no longer work for the newer drivers also?

[James Freeman]

Are you 100% sure about this? On my system I do not have an option for GPU in that box for my plasma display (driver version 331.65).
I only have that choice on the primary monitor display. Also, how does that affect custom resolutions.

I set to 0-255 by altering the INF before install. Does this no longer work for the newer drivers also?

No, sorry for that, I take that back.
Apparently its more complicated than that.

The Nvidia Range changes does not work with my setup (3770K, GTX660, v327.23).


Some findings:

* Choosing resolutions from the "Ultra HD,HD,SD" section is fixed to 16-235 signal.
But it will still include all the information without clipping it (only squeezing it to 16-235).

* Choosing resolutions from the "PC" section will send 0-255.
This is what we want, but sometimes not all wanted resolutions are available here.
We can create custom resolution to force it to PC range 0-255.


HDMI can send 2 types digital formats YCbCr444 and RGB.
They are both lossless.

RGB behaves like a VGA cable would (255,255,255).
YCbCr sends Chroma and Luma separately (video signal).


Digital Format (RGB, YCbCr444) & Resolution Section (HD, PC)
Together will determine the behavior and range of the HDMI signal.

1. UHD,HD,SD Resolutions + RGB format.
This option will send 0-255 signal to the TV, but squeeze all the information to 16-235 range*.
So the actual TV black point is at 0, but the signal blacks are at 16 and will look brighter (no BTB & WTW).
Not a useful option, In fact this is what we are complaining about.
* Unless you switch to 0-255 mode with madNvLevelsTweaker, then it becomes Full RGB signal like DVI/VGA.

2. UHD,HD,SD Resolutions + YCbCr444 format.
This option will send 0-255 signal to the TV, and not squeeze anything.
The actual black point is at 16 and white is at 235, so the picture looks good.
It will treat 0-15 as BTB & 235-255 as WTW.
So the picture looks perfect (like a standalone Blu-Ray Player would) and no information is lost.
Use your TVs Brightness & Contrast controls to calibrate the clipping points.

3. PC Resolutions, are always in 0-255 range like with DVI/VGA.
But you'll need a custom resolution if your desired resolution is not apparent in the PC resolutions section.


Note:
Most PC monitors only want to see RGB format (not YCbCr444) through the HDMI interface.
So you'll be forced to create Custom Resolution (if non existent in the PC section) to send 0-255 through HDMI.

But, most TVs will like the YCbCr444 format signal more,
as it exactly what a Blu-Ray/DVD player would send.


Hope this help.
I also hope I'm right this time. :cool:

[Werewolfy]

To force PC levels (0-255), there is a level tweaker included in the MadVr folder. It doesn't work anymore for you? On my config, it seems I have no problem with that.

And I think that madshi has always advised to output RGB and not YCbCr.

[James Freeman]

To force PC levels (0-255), there is a level tweaker included in the MadVr folder.

This one works.
Thank you.

And I think that madshi has always advised to output RGB and not YCbCr.

I can't quite agree about the RGB being better than YCbCr in a TV setup.

YCbCr has native Luma of 16-235, which we does not need to be converted and dithered to 0-255 with MadVR.
A TV (not a PC monitor) natively supports YCbCr through HDMI, so why not use that?

Setting YCbCr444 in Nvidia Contol Panel will output an HDMI signal exactly as a DVD/BR player would,
without the need to convert to RGB 0-255 or do any signal processing to the video content.

If one is connecting a PC through HDMI to watch a movie without leaving it as a permanent RGB monitor,
a YCbCr with 16-235 range in MadVR and calibrated Brightness & Contrast is the way to go (IMO).

If the TV is intended to replace a PC monitor, by all means go with RGB full range 0-255.

On the other hand, why would anyone want to use RGB 16-235 for anything?
Video (YCbCr) is 16-235, RGB is 0-255.
What RGB 16-235 is for, beats me.

[Werewolfy]

I can't quite agree about the RGB being better than YCbCr in a TV setup.

YCbCr has native Luma of 16-235, which we does not need to be converted and dithered to 0-255 with MadVR.
A TV (not a PC monitor) natively supports YCbCr through HDMI, so why not use that?

Setting YCbCr444 in Nvidia Contol Panel will output an HDMI signal exactly as a DVD/BR player would,
without the need to convert to RGB 0-255 or do any signal processing to the video content.

If one is connecting a PC through HDMI to watch a movie without leaving it as a permanent RGB monitor,
a YCbCr with 16-235 range in MadVR and calibrated Brightness & Contrast is the way to go (IMO).

If the TV is intended to replace a PC monitor, by all means go with RGB full range 0-255.

On the other hand, why would anyone want to use RGB 16-235 for anything?
Video (YCbCr) is 16-235, RGB is 0-255.
What RGB 16-235 is for, beats me.

Because MadVr outputs only in RGB ;) So if tou tell your video card to output in YCbCr, there is one more conversion needed.

[madshi]

Hi madshi, so I got a 2GB HD7850 updated to the latest 10/13 drivers(AMD didn't drop XP support despite of the hearsay) and when I open a movie in PotP, it opens in 960*540(as I set in my default PotP profile) but if I have FRC enabled, the video blinks like hell? Once I go FS and then windowed again, it doesn't blink anymore and it never does if FRC is disabled.

This problem doesn't happen if I have mVR's OSD enabled.

Refresh rate is properly detected as far as mVR's OSD can tell and this problem doesn't occur at all when I have a custom 3DLUT PS script enabled(a mirroring custom PS script gives some mirrored psychedelic effect), very strange...I'll hit you with log files in case you would like to investigate the problem but this is a non-blocking issue for me so no rush/biggy at all.

Also, the new rendering path is a GPU hog that goes all the way to 100% load but no biggy again as I've always been happy with the old one :)
The blinking sounds weird. Nobody else has ever reported that before, so it might be something specific to your PC, or maybe it's a driver issue or something. Don't know. The new rendering path works best on Windows 7 or Windows 8. XP really doesn't have any good rendering path, IMHO. The old FSE mode works "ok", but it's not really great.

I was wondering, is there a reason why in this kind of scene it debands so little? There is so little difference between debanding off and my current medium preset (the low preset doesn't do anything) and I have to increase every values very (too) much to obtain a good result.
I'm not sure. Sometimes things look different to a computer algorithm, compared to what our eyes see. I can investigate if you want, but I'd need a video sample for that.

I'm not sure if the answer to this is obvious, but why does a calibration shift over a period of time? I calibrated my monitor's using madshi's method in the madVR and ArgyllCMS thread on AVS and the gamma and color temperature shifts a bit only after around a week of usage. Is this due to aging of the actual LCD crystals or the variability of power delivered by the LCD pixel board?
I doubt your monitor shifted that much in such a short time. On my computer monitor the calibration is different depending on which corner of the display I'm measuring. Maybe you measured a different part of your screen compared to last time. Or maybe it was something related to warm up. Don't know...

I have a plasma that can accept both limited and full range through HDMI via a receiver.

If I set the panel to accept full range signal I get the following results using AVSHD basic patterns.

PS3 set to 'Y Pb/Cb Pr/Cr' (superwhite off) with full output over HDMI (for reference my PS3 shares same HDMI input on panel through my AV receiver as my PC which runs madVR)
- black = 16
- white = 235
If I set the PS3 super white on.
- black = 16
- white = 242

Now for MadVR with PC set to output YCbCr444 or RGB fullrange.

MADVR with PC levels 0-255
- black = 16
- white = 235

MADVR with PC levels 16-235
- black = 2
- white = 253
Not sure I understand this. Where do you get the 16, 235, 2 and 253 numbers from?

The only way I can get WTW output via the PC and madVR is to use custom output levels in MadVR
Then something is wrong somewhere.

Something legendary is cooking in this thread right now.
What d'you mean?

i have problem with editing the setting after play video all is good but no have icon to edit,
When restart PC first play i have shans to edit but second play again have this error message, any from here to know how to fix ?
That's quite weird. I'd suggest that you re-download madVR to make sure that the madHcCtrl.exe is not damaged in any way.

I've stumbled upon another weird thing on my ATI. I've begun to code a more responsive OSD with IOsdServices::OsdSetRenderCallback. It works great.
Yet on this ATI computer, any call using the madVR d3d device fails with D3DERR_INVALIDCALL.
It only works when Exclusive mode is enabled AND "present several frames in advance" is disabled...
I've tried to revert back to older driver with the same result. I just don't know what I'm doing wrong...
Sounds weird. What calls do you make to the d3d device? Have you tried something simple like GetDeviceCaps()? Does that fail, too?

I have tested the Upscaling Algorithms.

I have found that Jic3+AR gives me artifacts in HD and lower Chromas, whether Jinc3 (no AR) or Lanczos+AR did not.
And Anti-Ringing IS important, so Jinc without AR is a No-Go.
I can reproduce the artifacts. Not sure why they occur, maybe I'll look at that later. In any case, testing with artificial computer graphics test patterns should always be taken with a pinch of salt. Real movie content never has a cyan pixel right next to a red pixel. Why not? Because in real movie content lines are almost never ever perfectly vertical or horizontal. If you had such a high pixel contrast (red pixel next to cyan) with diagonal or curved lines, there'd be tons of extreme aliasing artifacts. This never happens with filmed content, due to several reasons, lens MTF being one of them. Linear resampling algorithms were never written for computer graphics. They were made for filmed content. And for such content they work much better than they work for computer graphics. Still, testing with computer graphics may help to get an idea of how a resampling algorithm deals with specific problems. But the usefulness is limited. The real test is true movie content.

Those artifacts you're seeing in that test pattern, nobody has ever reported or seen them yet. Neither in another test pattern, nor in a movie. So I doubt it's a common issue. Probably nothing to worry about.

For image upscaling I have concluded that I'd rather have a little aliasing instead of lost detail.
Therefor Lanczos 3/4+AR wins here in every resolution especially lower ones like DVD.
Why would you think that you lose detail when using Jinc? You're not. That test pattern with the black/white pixel on/off dither patterns does not represent "detail". You never have such a pattern with filmed content. You may have such patterns sometimes in print magazines, but not with filmed content. Jinc does not lose any detail. It might not always be as sharp as Lanczos, but that's not due to lost detail, that's due to lower aliasing.

Look here for a comparison with a real movie (Monsters AG Blu-Ray) which shows why many users prefer Jinc:

http://www.avsforum.com/t/1442663/which-hd-would-be-sufficient-for-madvr/120#post_23643869

I see NO difference what so ever in real videos.
Differences can be very small. Of course the bigger the scaling factor, the bigger the difference. Try some SD anime content, then look at jaggies in curved/diagonal hair lines. That sometimes shows the differences in aliasing rather clearly.

Ever since re-installing windows 7 I'm having an issue switching from a 50 or 72hz display to a 60hz display. The composition rate is detected as 24 or 25 hz by madvr and there's a lot of dropped frames. The only thing that fixes it is to switch to another video renderer, which plays fine, then switching back to madvr. Things that haven't worked; opening multiple madvr sessions, closing and reopening the player, reinstalling madvr, resetting madvr defaults. This used to happen rarely on old windows 7 installation but now it happens every time. Does anyone else experience this? Any other ideas to try?
Do you let madVR change display modes or do you use some other software? The composition rate is often not properly updated by other display mode changers, while madVR can usually do it properly.

YCbCr has native Luma of 16-235, which we does not need to be converted and dithered to 0-255 with MadVR.
A TV (not a PC monitor) natively supports YCbCr through HDMI, so why not use that?
Because Windows itself thinks and works in RGB. If you tell your GPU to output YCbCr, it will take the renderer's RGB output, convert it to YCbCr behind the renderer's back and then output it via HDMI. This is really bad because the GPU is likely to do the conversion is low quality.

Setting YCbCr444 in Nvidia Contol Panel will output an HDMI signal exactly as a DVD/BR player would,
without the need to convert to RGB 0-255 or do any signal processing to the video content.
Nope. Setting YCbCr444 in Nvidia Control Panel means that still everything is rendered in RGB, but the final RGB output will then be converted to YCbCr by the GPU driver.

There's one exception: When using hardware overlay, depending on GPU, driver, OS and renderer, it might eventually be possible to output untouched YCbCr. But e.g. AMD does not even support hardware overlay. And most renderers don't use hardware overlay at all. And I kinda doubt that the GPU drivers really support untouched YCbCr output properly.

On the other hand, why would anyone want to use RGB 16-235 for anything?
Video (YCbCr) is 16-235, RGB is 0-255.
What RGB 16-235 is for, beats me.
Computer monitors usually expect RGB 0-255 by default. However, consumer eletronic TVs and projectors are different. If you feed them with RGB, they usually expect 16-235 by default. Some TVs can be switched to expect RGB 0-255, but not all support this.

[leeperry]

The blinking sounds weird. Nobody else has ever reported that before, so it might be something specific to your PC, or maybe it's a driver issue or something. Don't know. The new rendering path works best on Windows 7 or Windows 8. XP really doesn't have any good rendering path, IMHO. The old FSE mode works "ok", but it's not really great.
OK, no biggy at all. I actually managed to make the AMD driver BSOD when rolling PS scripts so I've now created different profiles in PotP and roll profiles instead, so far so good :)

Anyway, I finally have a fully working rig with a 0.04cd/m² black and 3500:1 CR 4:4:4 A-MVA Sammy LCD TV, it simply looks out of this world with J3AR scaling and FRC :cool:

I would have prefered that you add a TOTL denoiser for upscaled SD instead of debanding as the latter would appear to always kill small details, but I reckon that some videos can look quite posterized sometimes so be it...I'm sure we'll all find a good use for it on otherwise hopeless videos and I'll keep looking for a denoising PS script.

OK got it, I've installed W7 in a VM, it looks quite good so I'll keep testing....at least no crazy tiles, skydrive integration and all that bs I'm really not ready to deal with.

[turbojet]

Ever since re-installing windows 7 I'm having an issue switching from a 50 or 72hz display to a 60hz display. The composition rate is detected as 24 or 25 hz by madvr and there's a lot of dropped frames. The only thing that fixes it is to switch to another video renderer, which plays fine, then switching back to madvr. Things that haven't worked; opening multiple madvr sessions, closing and reopening the player, reinstalling madvr, resetting madvr defaults. This used to happen rarely on old windows 7 installation but now it happens every time. Does anyone else experience this? Any other ideas to try?
Do you let madVR change display modes or do you use some other software? The composition rate is often not properly updated by other display mode changers, while madVR can usually do it properly.

I let madvr change at playback start but not at end. 80%+ of what's watched is 24fps. Changing it to 60hz at the end does fix the issue but another way I found is restarting dwm. Unfortunately I couldn't find a way to do it without admin or logging off/on. One way it could be done is calling sc stop/start with a program that stores admin credentials but that's a security risk, although it might be what I end up doing. How does madvr disable/enable composition?

Madvr could resolve the issue by restarting composition when display and composition rates do not match. Is this something that sounds reasonable?

[Razoola]

Not sure I understand this. Where do you get the 16, 235, 2 and 253 numbers from?


Then something is wrong somewhere.


Sorry about that. The numbers represent the physical colour bar I see on my display as black or white using the test pattern I mentioned (on the AVS-HD disk).

Like you say from my findings something some thing is going wrong somewhere but I'm not saying its the fault of madVR. I'm was just trying to point out to people that sometimes these issues are totally non madVR related.

That still leaves the question of how you want the pattern in question to look.

Some people say you want black at bar 16, white at 235 while I've seen others say you want black at 16 and white at about 20% over 235 (around 242). The calibration videos included say you want white right up to 254 if you can (see from 20:30 in this video http://revision3.com/hdnation/calibratepart2#rev3Player). Everything I have heard agree you want black at 16 but there is a large difference in opinions at where you want white to be. That is why I included ps3 info I get.

By playing with my setting (TV, gfx card and madVR) I can achieve all three of those situations without setting anything custom in madVR or making the picture washed out as a result. But to be honest I don't know which is the best. I can see the idea that black at 16, white at 235 is the spec. At the same time white over 235 means a larger color scale and can result in less white clipping.

[sebdelsol]

Sounds weird. What calls do you make to the d3d device? Have you tried something simple like GetDeviceCaps()? Does that fail, too?


Sorry, It was the CreateVertexBuffer call that failed actually. I should check more thoroughly ALL the return values :)
It works fine, since I've put all the ressources in a default pool instead of a managed one.
Only the drawText call fails, so I'll have to handle font rendering on my own it seems.
I guess that madVR uses a IDirect3DDevice9Ex ?

[James Freeman]

I can reproduce the artifacts. Not sure why they occur, maybe I'll look at that later.

Those artifacts you're seeing in that test pattern, nobody has ever reported or seen them yet. Neither in another test pattern, nor in a movie. So I doubt it's a common issue. Probably nothing to worry about.

Please try this video to see lots of artifacts with Jinc (Chroma Upscaling):
http://www.mediafire.com/download/wsinz06kt6xnfhw/color_bar_tests.mkv

Its the same pattern but compressed differently.
I found this one on the AVSForum board.

Jinc does not lose any detail. It might not always be as sharp as Lanczos, but that's not due to lost detail, that's due to lower aliasing.

Yes, I understand this.

But I still prefer how the "four way" look compared to "circular".
Jinc is too smooth.

[madshi]

I would have prefered that you add a TOTL denoiser for upscaled SD instead of debanding as the latter would appear to always kill small details, but I reckon that some videos can look quite posterized sometimes so be it...
Have you actually tested the madVR debanding algorithm yet? It should not kill detail (in any way worth mentioning) with the current "low" settings we've discussed in this thread.

I'm not a fan of noise/grain reduction at all. So that will probably be the very last algorithm I'll ever look into (if ever at all). If you think that debanding kills detail then what does noise reduction do? It's a hundred times more destructive to detail than (proper) debanding is, IMHO.

I let madvr change at playback start but not at end. 80%+ of what's watched is 24fps. Changing it to 60hz at the end does fix the issue but another way I found is restarting dwm. Unfortunately I couldn't find a way to do it without admin or logging off/on. One way it could be done is calling sc stop/start with a program that stores admin credentials but that's a security risk, although it might be what I end up doing. How does madvr disable/enable composition?

Madvr could resolve the issue by restarting composition when display and composition rates do not match. Is this something that sounds reasonable?
I'm already disabling and reenabling composition after every display mode change. Are you sure that there's no other software that changes display modes without you noticing? E.g. Reclock or MPC-HC?

Like you say from my findings something some thing is going wrong somewhere but I'm not saying its the fault of madVR.
It is very probably not the fault of madVR. Try bypassing your AVR. And make sure your GPU is really set to 0-255. You'll need to trick especially NVidia and Intel GPUs into 0-255 mode. They don't do that by default.

That still leaves the question of how you want the pattern in question to look.

Some people say you want black at bar 16, white at 235 while I've seen others say you want black at 16 and white at about 20% over 235 (around 242). The calibration videos included say you want white right up to 254 if you can
There's been a long thread on AVSForum about this. The general consensus seems to be that most Blu-Rays end (with true data) at 235, but some broadcasts seem to go up to 254. The specs call for 16-235, so that's what I would recommend to use. Of course you can go higher than 235 if you're afraid of clipping white, but that will come at a noticeable loss of brightness. I'd rather calibrate the display to do its best with proper material. Of course that's only my 2 cents.

It seems that madVR use a IDirect3DDevice9Ex ?
Yes, unless you activate the old FSE rendering path ("present several frames in advance" unchecked), then I have to use IDirect3DDevice9, because the old path doesn't work with IDirect3DDevice9Ex.

But I still prefer how the "four way" look compared to "circular".
Jinc is too smooth.
I can understand if you like the small sharpness advantage Lanczos has over Jinc. But there's really no such thing as "too smooth". Nature is smooth. You won't find aliasing in nature. Aliasing is an artifact of digital video / image processing, and there's no good side to it.

[leeperry]

Have you actually tested the madVR debanding algorithm yet? It should not kill detail (in any way worth mentioning) with the current "low" settings we've discussed in this thread.

I'm not a fan of noise/grain reduction at all. So that will probably be the very last algorithm I'll ever look into (if ever at all). If you think that debanding kills detail then what does noise reduction do? It's a hundred times more destructive to detail than (proper) debanding is, IMHO.

My point is that J3AR is great for upscaling SD but this level of accuracy requires a 4:4:4 display and then noise becomes a large problem IME, be it with computer monitors or TV's that disable their 4:2:2 built-in video processing.

I've found a few denoising PS scripts but all they really do is blur the picture, no more noise heh. I would entirely be willing to compromise small details for MPEG deblocking or VHS noise suppression. Sammy and especially Hitachi TV's do it really well IMHO but of course it's all 4:2:2.

I'm colorblind so judging on deposterizing is not an easy task for me and I'll give you that I've only looked at the worst case scenario screenshots posted in this thread, I'm sure the optimal settings you'll all agree upon will be well worth playing around with :)

The general consensus seems to be that most Blu-Rays end (with true data) at 235, but some broadcasts seem to go up to 254.
FWIW, Seb.26 implemented some sort of dynamic levels code in ffdshow a while ago: http://thumbnails103.imagebam.com/28559/da7379285589240.jpg (http://www.imagebam.com/image/da7379285589240)

There are indeed cases when you can find valuable data in the BTB/WTW but this can also end up being bogus padding and ffdshow works in 8bit anyway.

I can understand if you like the small sharpness advantage Lanczos has over Jinc. But there's really no such thing as "too smooth". Nature is smooth. You won't find aliasing in nature. Aliasing is an artifact of digital video / image processing, and there's no good side to it.

Yep, J3AR looks so natural....neither blurry like Bicubic or "digital" like Lanczos, yet another one-way ticket but it's still a major bummer that no picture viewer offers anything better than Lanczos :(

I think MPC can open JPG's so I could force the sRGB gamut in mVR et voilà. I'll try to get it to open PNG, GIF, RAR and ZIP. All I would need is a way to go between "1:1 scale" and "shrink to fit", which is not currently possible I think..

[James Freeman]

Another Chroma Upsampling test video to test which upsampler does a better job.

There are 2 versions inside.

1. 4:4:4, no chroma samplig at all (perfect RGB).
2. 4:2:0, a typical video chroma subsampling (BR/DVD).

For the 1080p Version:
Set MPC like so to keep 1:1 pixel ratio:
View -> Video Frame -> Normal Size.

For the DVD, set to Default "Touch Window From Inside"
and stretch the window to fullscreen.

Play the 4:2:0 video and pause it anywhere.
Now play with the "Chroma Upscaling" options.

Tell me what difference you see with Jinc, Lanczos, Bilinear, and Nearest Neighbor.

[LilScrappy]

please for advice, what is the best options for Exclusive Mode Settings - ( how many video frames shall be presented in advance 1-16 ? ) when is disabled Smooth Motion !
or when i use 24hz for my TV no matters what is Smooth Motion on or off is not functional when is play movie ?

[turbojet]

I let madvr change at playback start but not at end. 80%+ of what's watched is 24fps. Changing it to 60hz at the end does fix the issue but another way I found is restarting dwm. Unfortunately I couldn't find a way to do it without admin or logging off/on. One way it could be done is calling sc stop/start with a program that stores admin credentials but that's a security risk, although it might be what I end up doing. How does madvr disable/enable composition?

Madvr could resolve the issue by restarting composition when display and composition rates do not match. Is this something that sounds reasonable?

I'm already disabling and reenabling composition after every display mode change. Are you sure that there's no other software that changes display modes without you noticing? E.g. Reclock or MPC-HC?

It's not the display mode changing, it's the display itself. Going from a 72/50hz monitor to a 60hz tv and vice versa. Reclock isn't installed and players display mode changer is disabled. Changing display modes of the monitor works fine with madvr.

I was able to come up with a solution with very little to no security risk with a batch file:
DisplaySwitch.exe /internal ::or /external for TV
runasspc /cryptfile:"dwm.stop.spc" /quiet
runasspc /cryptfile:"dwm.start.spc" /quiet

dwm.start/stop.spc is an encrypted file containing admin credentials and sc start/stop uxsms.

[dansrfe]

Why is the composition rate set to the refresh rate of the primary display? Is this a flaw in Aero?

[James Freeman]

Although not the place, but I am happy to announce that the latest MPC-HC Nighty (LAV Filters 58.2.105) supports HEVC/H.265 & VP9 and works perfectly with MadVR.

A new era has began.

EDIT:
LAV 0.59 is out few minutes ago.

[Werewolfy]

I'm not sure. Sometimes things look different to a computer algorithm, compared to what our eyes see. I can investigate if you want, but I'd need a video sample for that.


Here's the video sample ;) http://www.mediafire.com/download/8joolhjv7mw9u9h/%5BDB%5D_Naruto_217_Sub_French_v2+fullanime.free.fr.avi


I'm not a fan of noise/grain reduction at all. So that will probably be the very last algorithm I'll ever look into (if ever at all). If you think that debanding kills detail then what does noise reduction do? It's a hundred times more destructive to detail than (proper) debanding is, IMHO.



I'm using FluxSmooth for the moment and with a good combination of spatial (value = 2) and temporal (value = 4) denoising, it doesn't destroy details and can be very useful with DVD, bad videos and even some mkv. It's like all postprocessing filters, it needs to be well configured. FluxSmooth has also the advantage to try to preserve the grain and just reduce the noise. I'd really like that one day you make a good denoising filter. I don't like AviSynth it's very buggy sometimes.

Another Chroma Upsampling test video to test which upsampler does a better job.

There are 2 versions inside.

1. 4:4:4, no chroma samplig at all (perfect RGB).
2. 4:2:0, a typical video chroma subsampling (BR/DVD).

For the 1080p Version:
Set MPC like so to keep 1:1 pixel ratio:
View -> Video Frame -> Normal Size.

For the DVD, set to Default "Touch Window From Inside"
and stretch the window to fullscreen.

Play the 4:2:0 video and pause it anywhere.
Now play with the "Chroma Upscaling" options.

Tell me what difference you see with Jinc, Lanczos, Bilinear, and Nearest Neighbor.
Nearest Neighbor has a lot of aliasing, Bilinear is a little too soft. I can't really see a difference between Bicubic, Lanczos and Jinc. I don't think that using Lanczos for chroma is a good idea. If you don't have the power to use Jinc, use Bicubic 75 + AR. Some people recommend to use Softcubic, it's a mistake, it's really too blur.

I was wondering, why does your video in 4:4:4 crash all the times with Ffdshow? It doesn't support 4:4:4? I use it only for some postprocessing filters.

[detmek]

please for advice, what is the best options for Exclusive Mode Settings - ( how many video frames shall be presented in advance 1-16 ? ) when is disabled Smooth Motion !
or when i use 24hz for my TV no matters what is Smooth Motion on or off is not functional when is play movie ?

1. Don't ask for best as it is against forum rules.
2. Don't change default settings unless you have problems like dropped or delayed frames.

[James Freeman]

.....I can't really see a difference between Bicubic, Lanczos and Jinc.

My thoughts exactly.

Anything above Bicubic+AR (or even Bilinear) for chroma upscaling is an overkill.
Although I still use Lanczos3+AR.

EDIT:
I added a more comprehensive Resolution Upscaling test in this post:
http://forum.doom9.org/showpost.php?p=1651152&postcount=20688

As you can see the difference is small to indistinguishable, except when the pixels are aligned in a
perfect diagonal (pixel by pixel), only then I see Jinc is doing something positive.
In all other diagonals Jinc just blurring them.

I was wondering, why does your video in 4:4:4 crash all the times with Ffdshow?

I have no idea.

I used the most basic settings in x264 for converting:
"x264.exe -o output.mkv input.mp4"

[Devrim]

Anyone else having issues with this sample with madvr? I just get a black screen.
http://forum.doom9.org/showthread.php?p=1651083#post1651083

In EVR it plays without issue

[huhn]

EDIT:
I added a more comprehensive Resolution Upscaling test in this post:
http://forum.doom9.org/showpost.php?...ostcount=20688

As you can see the difference is small to indistinguishable, except when the pixels are aligned in a
perfect diagonal (pixel by pixel), only then I see Jinc is doing something positive.
In all other diagonals Jinc just blurring them.


bilinear
http://i46.img-up.net/bilineare1fb.png
jinc 3 ar
http://v87.img-up.net/jinc3ara146.png

just have a look at the red 29 in the top right is better to read with jinc 3 ar

and the red border in the middle bottom is way better too

[James Freeman]

Anyone else having issues with this sample with madvr? I just get a black screen.
http://forum.doom9.org/showthread.php?p=1651083#post1651083

In EVR it plays without issue

Yeah.

Edit:
Try to disable Smooth Motion or dithering or change the Image Downscaling.

[sneaker_ger]

No problem with that sample here.
Radeon HD5850, Cat 13.11 Beta 8, Win7 x64
Laczos 3 Chroma Upscaling(AR), Catmull-Rom Downscaling(AR, LL) to 1920x1080 rectangle, dithering on

[James Freeman]

just have a look at the red 29 in the top right is better to read with jinc 3 ar

and the red border in the middle bottom is way better too

Looks like you comparing Chroma Upscaling not Image Upscaling.
It does look better compared to Bilinear.

But I still can't see difference between Bicubic and Jinc in Chroma Upscaling.

IMO,
Jinc for Image Upscaling is way too soft and hogs too much resources, Bicubic will do just fine and nothing beyond is visible OR needed.
For Chroma, again nothing beyond Bicubic is needed.

[huhn]

Anyone else having issues with this sample with madvr? I just get a black screen.
http://forum.doom9.org/showthread.php?p=1651083#post1651083

In EVR it plays without issue

works fine but the video kills my phenom 2 x4

[Chumbo]

The blinking sounds weird. Nobody else has ever reported that before, so it might be something specific to your PC, or maybe it's a driver issue or something. Don't know. The new rendering path works best on Windows 7 or Windows 8. XP really doesn't have any good rendering path, IMHO. The old FSE mode works "ok", but it's not really great.
Funny enough, I just ran into this strange "blinking" yesterday (only in windowed mode) and was about to post for some help but delved into the madVR settings and found a solution under scaling algorithms. Note that this is NOT an issue in full-screen exclusive mode.

This only happens when image upscaling/downscaling are set to use any of the options under processing done by custom pixel shader code (unaffected by chroma upscaling setting). Once I set these to use an options under either processing done by GPU texture units or processing done by GPU video logic windowed mode works fine.

My laptop uses an nVIDIA GeForce GT 635M with the latest 331.65 driver. Not sure if this is a driver issue or not.

[Ver Greeneyes]

From what I've heard a lot of people have reported problems with shaders and recent drivers on Nvidia cards (not madVR specific), so that might be related. Haven't seen any problems with madVR on my GeForce GTX 580 though.

[turbojet]

Why is the composition rate set to the refresh rate of the primary display? Is this a flaw in Aero?

I noticed this awhile back, can't remember what madshi said about it though. Not sure if aero is, or ever will be, capable of dealing with different refresh rates. Overlay and FSE work fine however. EVR and/or disabling composition fails.

Anyone else having issues with this sample with madvr? I just get a black screen.
http://forum.doom9.org/showthread.php?p=1651083#post1651083

In EVR it plays without issue

Try lowering GPU queue, going from 8 to 4 with 1GB vram worked for me.