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 3rd December 2016, 21:32   #41121  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,645
Madshi, I was looking over your anti-ringing thread because of Sp00kyFox's enquiry about the availability of your anti-ringing algo and noticed Tritical mentioning NNEDI4.

Did you ever evaluate it? I'd be interested in seeing the results of the modified prescreener. I guess it be fairly trivial to add given there's no major changes to NNEDI3. Tritical preferred the results of his newer version.

Never mind, I see he never released it.

Last edited by ryrynz; 4th December 2016 at 06:07.
ryrynz is offline   Reply With Quote
Old 3rd December 2016, 21:38   #41122  |  Link
AngelGraves13
Registered User
 
Join Date: Dec 2010
Posts: 243
Quote:
Originally Posted by madshi View Post
JFYI, have you tried using "low" quality downscaling?
I'm not a fan of bicubic. I like "soft" downscaling, but it seems you're going for a sharp downscaler with aliasing which then requires softening of edges. I guess everyone has their own taste.

Maybe for downscaling from NGU, you can add a "Soft" option which uses Jinc or something similar.
AngelGraves13 is offline   Reply With Quote
Old 3rd December 2016, 22:24   #41123  |  Link
fedpul
Registered User
 
Join Date: Feb 2014
Posts: 94
Quote:
Originally Posted by AngelGraves13 View Post
I'm not a fan of bicubic. I like "soft" downscaling, but it seems you're going for a sharp downscaler with aliasing which then requires softening of edges. I guess everyone has their own taste.

Maybe for downscaling from NGU, you can add a "Soft" option which uses Jinc or something similar.
I do not consider Jinc as a soft downscaler, it is very sharp, almost as sharp as SSIM or Bicubic 150. If you want something soft I will advice you to use Catmull Rom (Bicubic 50) or Softcubic (but be prepared to lose detail!)
fedpul is offline   Reply With Quote
Old 3rd December 2016, 23:17   #41124  |  Link
wiFFy
Registered User
 
Join Date: Oct 2012
Posts: 6
Quote:
Originally Posted by leeperry View Post
It's just that we've been playing around with all those settings for a long while now and we all got well polished profiles that make us plenty happy. Also, either you feed a very sharp picture to a soft downscaler or the opposite and settings such AR and LL downscaling are more a matter of personal taste and display dependent than anything else.

Anyway the what catastrophe has already been quite emotional lately so I'm not keen on giving up on SSIM2D100%LL+50%AB downscaling just yet huh, still rocking 0.91.1 and will be eagerly awaiting the next less crippled build please

This.

All these years with the "Judge it with your own eyes" isn't compatible with what looks like "Don't worry about your eyes, I already did the judging". Sane defaults produce indefinite value to any software, but they can't replace my individual preferences, that I have picked up over the years.

If you wish to tone down the settings, fine. Please just let us keep the advanced version of madVR in some form. I would be fine with hidden advanced settings ala ShowRenderSteps. Without it you could present as simplified UI as you like. With it you would get the current one.

It would require the user to come out and seek knowledge before they could start poking with the settings. I don't think the developer has any more responsibility to protect the user from themselves, if they wish to use some sort of harmful settings.

I would even rather have no UI for these advanced settings and use a plain text config file, than not have these advanced settings at all.

What sort of UI you decide to use in madVR 1.0 is up to you. I trust that you figure it all out by then. I just don't want to wait until madVR 1.0 to get the advanced settings back.

-EDIT: One idea for making madVR more newbie friendly: Tooltips. They are great for giving concentrated information for a new user without blowing up the UI. There is a currently a huge cliff between blindly staring at the settings with no explanation and googling a Madvr-guide. And trying to judge which guide to choose and trust. Or starting to browse a 2000 page thread on doom9 for information with even more opinions to choose from.

-EDIT2:More random thoughts for more newbie friendly madVR: Easy mode (=No learning required) Few presets (~3?) to choose from. Nothing to learn, nothing to ask. Services new users. Completely NGU-based + performance option(s) for when NGU won't do.

Advanced mode (=Current mode) What's new? Well, it's hidden, for starters. You can't stab yourself with a knife you can't find. Trying to make learning madVR easy, instead of taking out the need to learn.
-tooltips. Provides information without ever leaving the madVR UI. Possibly clickable, taking you to madVR Options Explained or equivalent.
-About madVR-tab: Clickable links to officially sanctioned (not necessarily made my madshi)documentation. *madVR Options Explained *madVR guide *madVR bug tracker *this thread *wiki (to be written, by users, to users) *button for activating profiles with a few premade ones (SD/720p/1080p)

Last edited by wiFFy; 4th December 2016 at 01:15.
wiFFy is offline   Reply With Quote
Old 4th December 2016, 01:10   #41125  |  Link
AngelGraves13
Registered User
 
Join Date: Dec 2010
Posts: 243
Quote:
Originally Posted by fedpul View Post
I do not consider Jinc as a soft downscaler, it is very sharp, almost as sharp as SSIM or Bicubic 150. If you want something soft I will advice you to use Catmull Rom (Bicubic 50) or Softcubic (but be prepared to lose detail!)
It's soft because it has no aliasing and minimal ringing. It doesn't need a lot of adjustments. And no, I don't think anyone here would consider Jinc to be sharp.

Last edited by AngelGraves13; 4th December 2016 at 05:36.
AngelGraves13 is offline   Reply With Quote
Old 4th December 2016, 07:42   #41126  |  Link
omarank
Registered User
 
Join Date: Nov 2011
Posts: 187
Quote:
Originally Posted by madshi View Post
Good to hear. Can you share a couple of screenshots which show why you like NGU low better than NGU high? Which scaling factor are we talking about?
When comparing screenshots/still images higher levels of NGU definitely look better (in many cases about the same). It is when watching a video I see that weirdness (depending upon the content). Anyway, I have taken screenshots of a scene which I think may give you an idea, although the difference between NGU low and NGU very high is rather too subtle with the stills. If you see the iron pipes of the gate in the foreground, they look too flat with NGU very high and more natural with NGU low. A Jinc scaled image is also there for reference. The scaling factor in this case is around 3, but I have been testing with scaling factors ranging from 2 to 3.

Quote:
Originally Posted by fedpul View Post
Results:

Option 1: It is from version 0.91.1 (Jinc AR)
Option 2: It is from version 0.91.3 (Bicubic 60 AR)

3 votes: 2 were right and 1 was wrong.

Because almost no one tried the challenge I will not be uploading another.

I was really hoping that more people get involved. I wanted to test if the ratio comes closer to 50/50 or not. And also it would have been useful to ask how did you compare the pics (for example if you use 1080p output or 4K), being honest I can't tell the difference between the two from normal view distance in a FHD 32" TV.
I can see the difference between the two images on a 50” FHD TV from about 5 feet distance. The important thing is that the two people who took the challenge (including me) replied correctly (the third one just queried about it), clearly indicating that the choice of scaling algorithm to be used after image doubling should be there as the differences can be seen.
omarank is offline   Reply With Quote
Old 4th December 2016, 07:58   #41127  |  Link
Magik Mark
Registered User
 
Join Date: Dec 2014
Posts: 666
Madshi

Experimented with NGU Med in chroma upscaling in 422 video. It took twice the processing power compared to 420. Is this expected? Big jump!


Sent from my iPhone using Tapatalk
__________________
Asus ProArt Z790 - 13th Gen Intel i9 - RTX 3080 - DDR5 64GB Predator - LG OLED C9 - Yamaha A3030 - Windows 11 x64 - PotPlayerr - Lav - MadVR
Magik Mark is offline   Reply With Quote
Old 4th December 2016, 08:33   #41128  |  Link
StinDaWg
Registered User
 
Join Date: Jan 2014
Posts: 216
Serious question... what type of video content are some of you guys using where you see such a difference between the different chroma algorithms?

I see zero difference between Bicubic 60AR and NGU Very High. I don't even know what I'm supposed to be looking for because the picture doesn't change at all. IMO it's a complete waste of gpu resources and I just use Bicubic 60AR on everything. Pretty sure madshi has said the same thing before. I just don't get it because I don't see any difference (unlike with luma upscaling where the difference is obvious). Is there only a difference when you zoom in 1000%? If so, then it's just placebo and who cares?
StinDaWg is offline   Reply With Quote
Old 4th December 2016, 08:54   #41129  |  Link
QBhd
QB the Slayer
 
QBhd's Avatar
 
Join Date: Feb 2011
Location: Toronto
Posts: 697
Well on my very old Plasma ("720p" which is 1024x768 rectangular pixels) different chroma algo's produce very noticeable differences in image sharpness and clarity. I like a sharp image and I can clearly see the difference between NNEDI3 32 and NNDEI3 64 in chroma upscaling... The chroma in image doubling is also noticeable, but to a lesser extant. Bicubic 60 in this context is far to soft. I don't know if it is a Plasma thing, or a rectangular pixel thing, but either way, control of chroma algo's is very important to me.

QB
__________________
QBhd is offline   Reply With Quote
Old 4th December 2016, 09:05   #41130  |  Link
Backflash
Registered User
 
Join Date: Jan 2016
Posts: 52
Quote:
Originally Posted by StinDaWg View Post
Serious question... what type of video content are some of you guys using where you see such a difference between the different chroma algorithms?

I see zero difference between Bicubic 60AR and NGU Very High
It's really really subtle, I operate purely in greyscale and it's almost invisible for me for wast majority of sources. So I don't understand the complaining about chroma neither.
What screen do you use? IPS would reflect it better in theory, never tried TN.

Quote:
Originally Posted by QBhd View Post
Well on my very old Plasma ("720p" which is 1024x768 rectangular pixels) different chroma algo's produce very noticeable differences in image sharpness and clarity. I like a sharp image and I can clearly see the difference between NNEDI3 32 and NNDEI3 64 in chroma upscaling... The chroma in image doubling is also noticeable, but to a lesser extant. Bicubic 60 in this context is far to soft. I don't know if it is a Plasma thing, or a rectangular pixel thing, but either way, control of chroma algo's is very important to me.

QB
It's probably plasma thing, it works with color differently compared to LCD.

Last edited by Backflash; 4th December 2016 at 09:08.
Backflash is offline   Reply With Quote
Old 4th December 2016, 09:06   #41131  |  Link
rxracer
Registered User
 
Join Date: Jul 2015
Posts: 3
DX11 crash & 80070005 error

I recently updated both MadVR and Nvidia drivers to v0.91.3 and v376.09 respectively. It's brought about 2 issues. The first is the media player crashes if DX11 is used, and the second is the error - creating Direct3D device failed (80070005).

The 80070005 error only affects a small % of files. All my videos are 1080p mkv (AVC). Initially I thought it might be related to frame rate, but I found some that have the same 25fps frame rate and they work fine. I noticed a while ago that these videos that are now freezing take @7 seconds to render an image and begin playing. I completely reset MadVR, to no avail. I did eventually manage to stop the error by enabling 'Use a separate device for presentation'. I never had to do this before though.

Since it's not a crash I can't send you a bug report. I tried ctrl+alt+shift+pause break as suggested when someone else had the same error in 2012, but nothing happens and I can't see a log. Let me know what you need me to do to help you.
rxracer is offline   Reply With Quote
Old 4th December 2016, 09:10   #41132  |  Link
leeperry
Kid for Today
 
Join Date: Aug 2004
Posts: 3,477
Quote:
Originally Posted by StinDaWg View Post
what type of video content are some of you guys using where you see such a difference between the different chroma algorithms?
Nothing touches Recon-Soft+chroma SR@2 on my 3500:1 AMVA UE32F5000@60Hz with BFI enabled and FRC in mVR from a 80 cm viewing distance, I mostly use 24p Hollywood movies and picture looks much more natural such as faces or landscapes. Last time I checked worst offender to my eyes was NNEDI3 for chroma, picture looked digital and artificial as can be.

Last edited by leeperry; 4th December 2016 at 09:13.
leeperry is offline   Reply With Quote
Old 4th December 2016, 09:15   #41133  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,645
Quote:
Originally Posted by QBhd View Post
I can clearly see the difference between NNEDI3 32 and NNDEI3 64 in chroma upscaling...
I would like to challenge you for an A/B compare.

I've done the comparisons myself and the pixel changes are.. basically impossible to notice.

Please send me five samples of your favorite content, whatever you like, I'll upscale a number of shots and we'll put your world class sight and equipment to the test.

Quote:
Originally Posted by Backflash View Post
It's probably plasma thing, it works with color differently compared to LCD.
As a plasma owner, it's actually harder to differentiate the differences due to the pixel noise.

Last edited by ryrynz; 4th December 2016 at 09:20.
ryrynz is offline   Reply With Quote
Old 4th December 2016, 10:59   #41134  |  Link
StinDaWg
Registered User
 
Join Date: Jan 2014
Posts: 216
Quote:
Originally Posted by QBhd View Post
Well on my very old Plasma ("720p" which is 1024x768 rectangular pixels) different chroma algo's produce very noticeable differences in image sharpness and clarity. I like a sharp image and I can clearly see the difference between NNEDI3 32 and NNDEI3 64 in chroma upscaling... The chroma in image doubling is also noticeable, but to a lesser extant. Bicubic 60 in this context is far to soft. I don't know if it is a Plasma thing, or a rectangular pixel thing, but either way, control of chroma algo's is very important to me.

QB
I have a 60" 1080p plasma. Looking at it with my nose almost up to the screen I don't see any difference between any of them.

I haven't looked at test patterns. This is just regular videos.

Last edited by StinDaWg; 4th December 2016 at 11:06.
StinDaWg is offline   Reply With Quote
Old 4th December 2016, 11:52   #41135  |  Link
Sunset1982
Registered User
 
Join Date: Sep 2014
Posts: 280
Quote:
Originally Posted by rxracer View Post
I recently updated both MadVR and Nvidia drivers to v0.91.3 and v376.09 respectively. It's brought about 2 issues. The first is the media player crashes if DX11 is used, and the second is the error - creating Direct3D device failed (80070005).

The 80070005 error only affects a small % of files. All my videos are 1080p mkv (AVC). Initially I thought it might be related to frame rate, but I found some that have the same 25fps frame rate and they work fine. I noticed a while ago that these videos that are now freezing take @7 seconds to render an image and begin playing. I completely reset MadVR, to no avail. I did eventually manage to stop the error by enabling 'Use a separate device for presentation'. I never had to do this before though.

Since it's not a crash I can't send you a bug report. I tried ctrl+alt+shift+pause break as suggested when someone else had the same error in 2012, but nothing happens and I can't see a log. Let me know what you need me to do to help you.
Did you set up 3D? There is an old bug with nvidia where madvr crashes when d3d11 is used and 3d is set up in the driver. Just an idea to check out before you investigate it further...
__________________
Intel i5 6600, 16 GB DDR4, AMD Vega RX56 8 GB, Windows 10 x64, Kodi DS Player 17.6, MadVR (x64), LAV Filters (x64), XySubfilter .746 (x64)
LG 4K OLED (65C8D), Denon X-4200 AVR, Dali Zensor 5.1 Set
Sunset1982 is offline   Reply With Quote
Old 4th December 2016, 14:17   #41136  |  Link
rxracer
Registered User
 
Join Date: Jul 2015
Posts: 3
Quote:
Originally Posted by Sunset1982 View Post
Did you set up 3D? There is an old bug with nvidia where madvr crashes when d3d11 is used and 3d is set up in the driver. Just an idea to check out before you investigate it further...
No, 3D is always off in MadVR. I don't even install the Nvidia 3D drivers. Can't remember the last time I used 3D. And since the player now crashes with DX11 enabled I've got it disabled. Back to DX9 atm.

It's an odd one. Things are fine so long as 'use a separate device for presentation' is checked. I still get an initial delay of @7 seconds before playback, but that's been the case for a while. It's only with specific rips, so I may not have noticed it for a while if not for these few. I checked my Sonarr history and it's a group and source I don't recall grabbing before. Humans.S02E01.PROPER.1080p.AMZN.WEBRip.DD2.0.x264-KiNGS.

Since I've found a workaround this is mostly for Madshis' benefit if it's of consequence to him. I am curious to know the reason though. It's a shame MadVR doesn't crash so I can just send a report. I sent one a long time ago and he was really helpful. He seems to properly care about MadVR. I respect that.
rxracer is offline   Reply With Quote
Old 4th December 2016, 14:29   #41137  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by Telion View Post
Nope. And I tried in v0.91.1 the same settings that new presets are giving - everything is OK. In v0.91.3, when I switch back and forth between bicubic and NGU low for chroma on the fly while in NGU doubling mode and on pause, the picture immediately switches between a mess and a correct one accordingly.
That's really weird, especially because you seem to be the only one affected by this problem. Have you tried a different GPU driver version?

Quote:
Originally Posted by hannes69 View Post
Feedback for NGU watching lores/lowbitrate content:
Iīm watching snooker at the moment with Kodi + MPC HC + madVR. Iīm watching it with BBC iplayer addon and because Iīm living outside UK Iīm using a proxy (-> low bandwidth -> low video quality). BBC video material has a good quality already with low bandwidth, but it is blurry of course.
I tested all scaling algorithms, and I have to say: NGU is one of the most noticeable improvements in madVR for me!!!
Of course there were other great improvements like the display switcher, dithering, smoothmotion, debanding, profiles and so on, but in regards of scaling NGU tops everything in my opinion.
Back to my example: the material I watched had 512x288 resolution and a bitrate of 544kbps (I consider this really low quality). I scaled it to my monitor resolution of 1680x1050.
What shall I say? NGU has a new dimension in scaling quality. BTW I can use NGU low up to NGU high with high quality downscaling with my very weak Radeon R7 250, thatīs really high bang for the buck...
With low quality scalers the mentioned video material is on the edge being watchable for me; by using NGU itīs absolutely normally watchable without thinking about the bad quality all the time...
NGU beats clearly NNEDI (and of course all other scalers) for me when watching this lowres content.
NGU low is sufficient sharp, NGU medium is on the edge of being too sharp (or there are artifacts or it looks somehow unnatural), NGU high is worse in this regard compared to medium. So I stay with medium at the moment.
I share the opinion that a scaler should produce "the real image" and not add something like sharpening additionally. So NGU low/medium seem to look more natural/unprocessed to me than NGU high.
Of course high quality content is another story (havenīt had the time to test that much).

NGU is an impressive improvement in scaling to me. Because:
I normally go mostly for judderfree playback (-> custom resolution / smoothmotion) and color accuracy (->calibrated monitor and beamer), scaling was for me a thing for microscope nerds using 800% magnifying and watching for single pixels. In normal usage scenarios I often didnīt see differences between several scalers, linear light, debanding, dithering and so on (sometimes I thought I saw something, most of the time it was placebo I think). NGU is a completely other story. I see CLEARLY the difference compared to other scalers, it CLEARLY wins in regards of quality and performance and I didnīt have to use 800% magnified pictures to find out about that.
That sounds pretty great!! I've put a LOT of work into NGU, so I'm happy to hear it pays off. JFMI: Could you maybe show a snooker screenshot comparing NGU vs whatever you were using before?

Generally, NGU handles soft images fine. However, blocking artifacts and especially mosquito noise throws it off.

Quote:
Originally Posted by ryrynz View Post
noticed Tritical mentioning NNEDI4.

Did you ever evaluate it? [...]

Never mind, I see he never released it.
Yep, he never released it. From the few images I had seen I didn't feel it was a noticeable improvement. He felt it was better, though.

Quote:
Originally Posted by omarank View Post
When comparing screenshots/still images higher levels of NGU definitely look better (in many cases about the same). It is when watching a video I see that weirdness (depending upon the content). Anyway, I have taken screenshots of a scene which I think may give you an idea, although the difference between NGU low and NGU very high is rather too subtle with the stills. If you see the iron pipes of the gate in the foreground, they look too flat with NGU very high and more natural with NGU low. A Jinc scaled image is also there for reference. The scaling factor in this case is around 3, but I have been testing with scaling factors ranging from 2 to 3.
Thanks. Hmmm... I'm not sure I can see what you mean about the pipes. They do look "different" somehow, but it's hard to say for me which is better.

Quote:
Originally Posted by Magik Mark View Post
Experimented with NGU Med in chroma upscaling in 422 video. It took twice the processing power compared to 420. Is this expected? Big jump!
422 is twice the pixels compared to 420, so that's expected. NGU cannot upscale in only one direction (X or Y) like NNEDI3 can. NGU always upscales both X and Y by design. So with 422 content NGU actually has to upscale X and Y to 200% and then downscale one dimension again. So it's a lot of extra work compared to 420.

Quote:
Originally Posted by StinDaWg View Post
Serious question... what type of video content are some of you guys using where you see such a difference between the different chroma algorithms?
It's only really visible in specific scenes. Look for scenes with lots of sharp red stuff, preferably on top of a dark/black background.

Quote:
Originally Posted by rxracer View Post
I recently updated both MadVR and Nvidia drivers to v0.91.3 and v376.09 respectively. It's brought about 2 issues. The first is the media player crashes if DX11 is used, and the second is the error - creating Direct3D device failed (80070005).
Probably caused by the driver upgrade. Maybe you can downgrade the driver?
madshi is offline   Reply With Quote
Old 4th December 2016, 14:34   #41138  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
madVR v0.91.4 released

http://madshi.net/madVR.zip

Code:
* redesigned "image upscaling" settings page once more
* super-xbr image doubling is available again
* fixed: downscaling anti-ringing checkbox was sometimes disabled
I hope you guys will like the new "image upscaling" settings page. I've compromised a little, gave you back some more control, but tried to keep it easy to understand for normal users.

I've dropped the option to let you choose with which exact scaling factor doubling gets activated. madVR now makes that decision itself. For NGU, doubling is activated as soon as any upscaling is needed, because NGU has some nice properties (e.g. removes ringing) which can be beneficial even when upscaling only by a very small factor. NNEDI3 and super-xbr are only activated if the scaling factor is 1.2x or higher. Of course you can use profiles to force any doubler to activate or deactivate in specific situations. If you've selected any of the doublers and they don't get activated, madVR now falls back to JincAR upscaling.
madshi is offline   Reply With Quote
Old 4th December 2016, 14:57   #41139  |  Link
QBhd
QB the Slayer
 
QBhd's Avatar
 
Join Date: Feb 2011
Location: Toronto
Posts: 697
I like the new layout, I just did a quick peak so I haven't really had a chance to use it yet, however first impressions are really good

QB
__________________
QBhd is offline   Reply With Quote
Old 4th December 2016, 15:03   #41140  |  Link
LoRd_MuldeR
Software Developer
 
LoRd_MuldeR's Avatar
 
Join Date: Jun 2005
Location: Last House on Slunk Street
Posts: 13,248
I wonder, when does the "<-- upscaling" option for NGU apply?

As far as I understand, NGU doubling is now always enabled when upscaling is needed (even for small factors). So, when upscaling is needed, we always get NGU doubling, possibly (usually) followed by "<-- downscaling".

But in what case is NGU "<-- upscaling" used?
__________________
Go to https://standforukraine.com/ to find legitimate Ukrainian Charities 🇺🇦✊

Last edited by LoRd_MuldeR; 4th December 2016 at 15:22.
LoRd_MuldeR 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 08:50.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.