Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Hardware & Software > Software players

Reply
 
Thread Tools Search this Thread Display Modes
Old 10th May 2016, 15:46   #37861  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,137
Quote:
Originally Posted by 70MM View Post
APPLY is greyed out so you cant activate the new feature.

Im not the only one to report this now.
I've not seen anybody else reporting this yet.

Do you have one madVR installation or multiple? Have you updated *all* the files on all PCs involved?

Quote:
Originally Posted by omarank View Post
Well, the application doesn’t require installation. You can just run the standalone exe file. Anyway, please see the screenshots in the below link:

http://www.mediafire.com/view/pd8924...Comparison.png

The application doesn’t resize so I am unable to show the complete table. But, I guess with the screenshots you will get a fair idea of the problem.
Yes, as expected. The application doesn't seem to understand that there are 2 different ways to create a linear lut. IIRC it can even vary depending on OS and GPU. In any case, madVR supports both and from where I'm standing there's nothing wrong with what madVR does.

Quote:
Originally Posted by omarank View Post
Nevcairiel was willing to output XYZ directly to madVR, when I made that request. He also said that it wouldn’t be too much work on his side. Maybe you can speak to him.

For testing, you may download the trial version of JPEG2000 Video Decoder. More information here.

By default it outputs XYZ (disguised as RGB). You can enable the XYZ to RGB conversion through the decoder settings. It’s output is limited to 8 bits only. Honestly, the ideal solution would be that LAV outputs XYZ to madVR.
Still waiting for RenderGuy2 to send me whatever he wanted to send me. I've put this on my to do list, but as I said, it'll be low priority.

Quote:
Originally Posted by omarank View Post
By the way, the file name tag “deringing=full|half” would enable/disable “reduce dark halos around bright edges, too” option, right?
Yes.

Quote:
Originally Posted by ryrynz View Post
The dehaloing is nice, but what are the chances of it being a little less soft/smudgy, I'd like to see it hopefully retain or "create" some kind of detail if possible. Can you add some grain and or
allow the algo to be a little more flexible in it's choices of color?
I suppose I could add grain. Color is not modified, only the luma channel is.

Quote:
Originally Posted by cork_OS View Post
New anti-ringing algo looks really great, but sometime needs even more strength. Sample: AR / noAR.
Well, the algo is tweaked for "normal" halos, not for extreme halos. I'm not sure if I could remove those. I don't plan any strength settings in the near future.

Quote:
Originally Posted by aufkrawall View Post
Unfortunately, it looks kinda weird after upscaling the American Dad cartoon example, the leftovers are much more obvious than the ringing in general.
Really? In the American Dad example, I didn't really notice any noticeable leftovers. It looks very clean to me, and noticeably better than without deringing. Can you show an upscaled American Dad screenshot with/without deringing which shows that deringing makes it look weird?
madshi is offline   Reply With Quote
Old 10th May 2016, 16:07   #37862  |  Link
regiregi22
Registered User
 
Join Date: May 2014
Posts: 20
Quote:
Originally Posted by madshi View Post
This was discussed AT LENGTH just a couple of pages ago.
I know, I have read those pages. But it has some cases where there was limitations I don't have. Like output devices that don't support extended ranges or dedicated HTPCs that don't care about messing up colors on desktop applications. There were shown quite some alternatives, each one with each drawbacks.

So that's why I am asking, I can assure you I have read them paying attention.

Thank you for your input anyway :-)
regiregi22 is offline   Reply With Quote
Old 10th May 2016, 16:22   #37863  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,708
Quote:
Originally Posted by madshi View Post
Really? In the American Dad example, I didn't really notice any noticeable leftovers. It looks very clean to me, and noticeably better than without deringing. Can you show an upscaled American Dad screenshot with/without deringing which shows that deringing makes it look weird?
NNEDI3 32 chroma scaling, NNEDI3 64 doubling, super-xbr AB 25% quadrupling, bicubic 150 AR GL downscaling, SuperRes 4 GL AB 25%

off:


on (recommended setting for Anime):
aufkrawall is offline   Reply With Quote
Old 10th May 2016, 16:30   #37864  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,137
Quote:
Originally Posted by regiregi22 View Post
I know, I have read those pages. But it has some cases where there was limitations I don't have. Like output devices that don't support extended ranges or dedicated HTPCs that don't care about messing up colors on desktop applications. There were shown quite some alternatives, each one with each drawbacks.

So that's why I am asking, I can assure you I have read them paying attention.
This is the most asked question ever, and really going on my nerves because of that. Here's a quote of my sum up, which I've posted ages ago:

http://forum.doom9.org/showpost.php?...ostcount=37690

I won't say any more about it.

Quote:
Originally Posted by aufkrawall View Post
NNEDI3 32 chroma scaling, NNEDI3 64 doubling, super-xbr AB 25% quadrupling, bicubic 150 AR GL downscaling, SuperRes 4 GL AB 25%
Oh well, yes there are some leftovers... I still prefer the deringed image, but I suppose it's a matter of taste/preferences...
madshi is offline   Reply With Quote
Old 10th May 2016, 16:32   #37865  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,708
Quote:
Originally Posted by madshi View Post
Oh well, yes there are some leftovers... I still prefer the deringed image, but I suppose it's a matter of taste/preferences...
I think the deringing filter works fine, but SuperRes introduces some of the ringing again in a nasty way.
Could it be that SuperRes doesn't compare the upscaled image with the source image which is already processed by deringing?
aufkrawall is offline   Reply With Quote
Old 10th May 2016, 16:34   #37866  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,137
Nah, SuperRes properly compares the image after deringing and before upscaling, that's not it, unfortunately.
madshi is offline   Reply With Quote
Old 10th May 2016, 16:43   #37867  |  Link
James Freeman
Registered User
 
Join Date: Sep 2013
Posts: 919
madshi, is there a way to ignore wrong ST.2086 metadata and use peak white was selected by user?
The official test video from Sony has completely useless metadata and madVR simply shows all white picture (0.5 nit peak white):

* The video plays perfectly fine on a Samsung HDR TV as reported by users.
It ignores the 2086 metadata completely, and just maps the bit values to luminance.

http://demo-uhd3d.com/fiche.php?cat=uhd&id=144
Quote:
Video
ID : 1
Format : HEVC
Format/Info : High Efficiency Video Coding
Format profile : Main 10@L5.1@High
Codec ID : hvc1
Codec ID/Info : High Efficiency Video Coding
Duration : 2mn 7s
Bit rate : 75.6 Mbps
Maximum bit rate : 123 Mbps
Width : 3 840 pixels
Height : 2 160 pixels
Display aspect ratio : 16:9
Frame rate mode : Constant
Frame rate : 59.940 (60000/1001) fps
Color space : YUV
Chroma subsampling : 4:2:0 (Type 2)
Bit depth : 10 bits
Bits/(Pixel*Frame) : 0.152
Stream size : 1.12 GiB (100%)
Encoded date : UTC 2016-02-03 07:59:49
Tagged date : UTC 2016-02-03 08:01:32
Color range : Limited
Color primaries : BT.2020
Transfer characteristics : SMPTE ST 2084
Matrix coefficients : BT.2020 non-constant
Mastering display color primar : R: x=1.000000 y=1.000000, G: x=1.000000 y=1.000000, B: x=1.000000 y=1.000000, White point: x=1.000000 y=1.000000
Mastering display luminance : min: 0.1000 cd/m2, max: 0.5000 cd/m2
Let the user select 1200nit as peak white ignoring the wrong 0.5nit 2086 metadata.
__________________
System: i7 3770K, GTX660, Win7 64bit, Panasonic ST60, Dell U2410.

Last edited by James Freeman; 10th May 2016 at 17:07.
James Freeman is offline   Reply With Quote
Old 10th May 2016, 16:45   #37868  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,708
Quote:
Originally Posted by madshi View Post
Nah, SuperRes properly compares the image after deringing and before upscaling, that's not it, unfortunately.
Too bad, your deringing filter seems to work wonders in general.
Especially with super-xbr, the difference in the cartoon example without SuperRes is enormous.
aufkrawall is offline   Reply With Quote
Old 10th May 2016, 16:50   #37869  |  Link
har3inger
Registered User
 
Join Date: Feb 2014
Posts: 139
Didn't Superres at 4 with AB introduce a bit of ringing anyway? Maybe try at superres 2 no AB or superres off to see if it's still bad. The sample aufkrawall posted seems noticeably worse than the one posted by madshi in his release post.
har3inger is offline   Reply With Quote
Old 10th May 2016, 16:53   #37870  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,708
SuperRes introduces ringing when it is in the source, especially with higher strength. You are right that it's not a problem with a strength of 1.
aufkrawall is offline   Reply With Quote
Old 10th May 2016, 17:05   #37871  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 5,722
@aufkrawall

can you try super XBR without anti bloat?

the artefects kind of look like the same issue i have with super XBR AB with crisp edges.
huhn is offline   Reply With Quote
Old 10th May 2016, 17:08   #37872  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,708
It seems to happen totally independent from the chosen upscaling algorithm, also with super-xbr or Jinc the ringing looks "aliased" with deringing + high SuperRes strength.
aufkrawall is offline   Reply With Quote
Old 10th May 2016, 17:31   #37873  |  Link
YxP
Registered User
 
Join Date: Oct 2012
Posts: 97
What a remarkable addition this deringing is. Ringing is my number one annoyance, much more so than mild banding or aliasing. Whatever leftovers you guys are talking about, they definitely are not as bad as ringing itself :P Great work!
YxP is offline   Reply With Quote
Old 10th May 2016, 17:39   #37874  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,708
Did you see my comparison above? The ringing in some areas of the image gets more distinct than without deringing because SuperRes with high strength reintroduces it in an aliased shape. Which is of course not the fault the deringing filter, but an unfortunate coincidence.
aufkrawall is offline   Reply With Quote
Old 10th May 2016, 17:44   #37875  |  Link
XMonarchY
Registered User
 
Join Date: Jan 2014
Posts: 488
Quote:
Originally Posted by nevcairiel View Post
Why don't you try and find out?
If you can't judge the quality for yourself, then you might as well not bother, because you won't see a difference then either.
I don't see much difference, but I also don't see that much of a difference between most of madVR settings. The image quality improvement that I obviously see only comes when all high-quality settings are selected and utilized. Hell, we both know that most image/chrome upscaling/downscaling method comparisons shown here require zooming in on particular specific parts of images because its very hard, if not impossible, to see the difference at native resolution / "with a naked eye" so-to-speak.

I know there was a general consensus that Scale in Linear Light provided the best results for Catmull-Rom Image Downscaling method, but I do not know if the same applies to the SSIM 2D 100% Image Downscaling method. What is the general consensus - should I use Scale in Linear Light for SSIM 2D 100% downscaling.
__________________
8700K @ 5Ghz | ASUS Z370 Hero X | Corsair 16GB @ 3200Mhz | RTX 2080 Ti @ 2100Mhz | Samsung 970 NVMe 250GB | WD Black 2TB | Corsair AX 850W | LG 32GK850G-B @ 165Hz | Xonar DGX | Windows 10 LTSC 1809
XMonarchY is offline   Reply With Quote
Old 10th May 2016, 17:51   #37876  |  Link
YxP
Registered User
 
Join Date: Oct 2012
Posts: 97
Quote:
Originally Posted by aufkrawall View Post
Did you see my comparison above? The ringing in some areas of the image gets more distinct than without deringing because SuperRes with high strength reintroduces it in an aliased shape. Which is of course not the fault the deringing filter, but an unfortunate coincidence.
Yes I did see it and you are correct of course. I'm just so impressed at the moment that maybe I need some time to start paying attention to the leftovers, as you call them. Personally I only use SR@1 or 2 because more will make my GPU fan go nuts.
YxP is offline   Reply With Quote
Old 10th May 2016, 18:02   #37877  |  Link
omarank
Registered User
 
Join Date: Nov 2011
Posts: 180
Quote:
Originally Posted by madshi View Post
Yes, as expected. The application doesn't seem to understand that there are 2 different ways to create a linear lut. IIRC it can even vary depending on OS and GPU. In any case, madVR supports both and from where I'm standing there's nothing wrong with what madVR does.
Ok. I have this basic understanding that all output from the OS is passed through the GPU LUT. If the LUT is linearized to unity, then all video data will be multiplied by 1 and the output will be unchanged by the LUT. In any other case, there will be floating point/ high bit depth values which the GPU will have to dither to the output bit depth.

Now it is interesting to learn that LUT can be set in two different ways and still remain linear. Could you please explain a bit how that works, if you don’t mind?

Quote:
Originally Posted by madshi View Post
Still waiting for RenderGuy2 to send me whatever he wanted to send me. I've put this on my to do list, but as I said, it'll be low priority.
Sure, thanks!

While we are at it, I will share some thoughts:

1. 3D DCP video is a frame sequential 48fps stream with alternate left eye and right eye frames for 24 fps content. If LAV video decoder just flags and sends the stereo frames in the same way as it does for MVC content, no special handling of 3D would be required in madVR (I believe).

2. Currently, the black bar detection algorithm doesn’t work for RGB content (and also YUV content, IIRC), so it wouldn’t work for XYZ as well without extra work on your side. However, you can use the “Y” of XYZ for black bar detection as you do for YCbCr content because “Y” in XYZ and YCbCr are very similar as both have lightness information.
omarank is offline   Reply With Quote
Old 10th May 2016, 18:21   #37878  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,708
Quote:
Originally Posted by XMonarchY View Post
I know there was a general consensus that Scale in Linear Light provided the best results for Catmull-Rom Image Downscaling method, but I do not know if the same applies to the SSIM 2D 100% Image Downscaling method. What is the general consensus - should I use Scale in Linear Light for SSIM 2D 100% downscaling.
I don't think there is a difference in "quality". LL downscaling should maintain the original brightness of the unscaled image while gamma light is darker. So imho it makes sense to use LL when there is just downscaling happening (imho it doesn't even matter which downscaling algorithm is used, LL is always brighter and GL a bit less aliased).

However, SuperRes also makes the image even brighter (in GL more than in LL, that's a little confusing since it's the opposite way around compared to image scaling) and must be fed with image in GL when SuperRes itself also works in GL to avoid aliasing introduction (and fed with LL when it works in LL).
So, it should be best to define script profiles to achieve LL downscaling when there is no upscaling with SuperRes and to use GL downscaling when SuperRes in GL is used (it should be used in GL since probably most content is that way).
aufkrawall is offline   Reply With Quote
Old 10th May 2016, 19:06   #37879  |  Link
AngelGraves13
Registered User
 
Join Date: Dec 2010
Posts: 239
SuperRes at 1 is the best option in my opinion. Anything more than 1 starts causing its own issues with the image. 4 usually looks bad unless you have a super clean image, and even then I'd say 3 should be the limit on SuperRes.
AngelGraves13 is offline   Reply With Quote
Old 10th May 2016, 19:15   #37880  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,708
Really depends on the source and target device.
I hardly see any artifacts with 1080p -> WQHD Jinc AR SL + SR 4, except with broken sources or small or bad fonts (which would also look pixelated on a 24" 1080p display with 1080p source).
aufkrawall is offline   Reply With Quote
Reply

Tags
direct compute, dithering, error diffusion, madvr, ngu, nnedi3, quality, renderer, scaling, uhd upscaling, upsampling

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 10:55.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2019, vBulletin Solutions Inc.