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 December 2016, 03:33   #41321  |  Link
pirlouy
_
 
Join Date: May 2008
Location: France
Posts: 692
Your impression is not enough. From what I see, Madshi reads scientific reviews to learn about new algorithms and progress in image improvment. When he makes a decision, there's a big chance it is the good one.
Don't forget the best image is not what you like the most, but the one which looks like the original the most. So we'll have to do ala Hydrogenaudio dictatorship: you need to provide proofs from your analyse with screenshots like this post for example. Madshi said he won't add option for one guy, so you have to convince him !
pirlouy is offline   Reply With Quote
Old 10th December 2016, 04:07   #41322  |  Link
Shiandow
Registered User
 
Join Date: Dec 2013
Posts: 753
Quote:
Originally Posted by pirlouy View Post
Your impression is not enough. From what I see, Madshi reads scientific reviews to learn about new algorithms and progress in image improvment. When he makes a decision, there's a big chance it is the good one.
Don't forget the best image is not what you like the most, but the one which looks like the original the most.
While I agree that there are some 'misguided' ways to process an image, and that by knowing more about the theory behind the methods you can make some better judgements on what kind of processing is and isn't a good idea, let's not get confused about the end goal. The goal is to get the best image, and this is purely subjective (this isn't up for discussion by the way, it says so in the forum rules ).

All these heuristics like downscaling an image and upscaling it again with different algorithms to see which is closest to the original, are just that, heuristics. They are useful for comparing algorithms, or even designing new ones, but they're no substitute for actually looking yourself.
Shiandow is offline   Reply With Quote
Old 10th December 2016, 05:29   #41323  |  Link
AngelGraves13
Registered User
 
Join Date: Dec 2010
Posts: 255
madshi, please add an option to DISABLE chroma doubling for those who don't want/need it in the drop-down list. I can run VeryHigh NGU, but the chroma doubling seems useless, even with the Normal (Bicubic 60) option. I'd just like to disable the chroma doubling altogether.

I'm going back to Jinc SR4 for Chroma and Upscaling. Looks just as good and has very low render times (15ms)

Last edited by AngelGraves13; 10th December 2016 at 05:49.
AngelGraves13 is offline   Reply With Quote
Old 10th December 2016, 05:37   #41324  |  Link
jkauff
Registered User
 
Join Date: Oct 2012
Location: Akron, OH
Posts: 491
Quote:
Originally Posted by Shiandow View Post
While I agree that there are some 'misguided' ways to process an image, and that by knowing more about the theory behind the methods you can make some better judgements on what kind of processing is and isn't a good idea, let's not get confused about the end goal. The goal is to get the best image, and this is purely subjective (this isn't up for discussion by the way, it says so in the forum rules ).

All these heuristics like downscaling an image and upscaling it again with different algorithms to see which is closest to the original, are just that, heuristics. They are useful for comparing algorithms, or even designing new ones, but they're no substitute for actually looking yourself.
Gospel from Shiandow. Should be a sticky.
jkauff is offline   Reply With Quote
Old 10th December 2016, 05:52   #41325  |  Link
AngelGraves13
Registered User
 
Join Date: Dec 2010
Posts: 255
Quote:
Originally Posted by ryrynz View Post
Really need some animated or real world content shot in 4:4:4..

It's easier to pick a winner when there's just one color like in above scenario but it may just be that more times than not NGU very high is the most accurate.
Some extra tests between the top contenders.

Super-xbr 100 AR vs NGU low
NGU Low vs NGU med
NGU med vs NGU high
NGU high vs NGU very high
NGU low vs NGU very high



NNEDI3 32 is quite nice and IMO looks better than NGU low but it's considerably more demanding at least on this HD4000, it's more demanding than NGU very high which makes it hard to recommend.
One wonders if perhaps it's time for NNEDI3 to retire from chroma duties.. That asks for one more I guess..

NNEDI3 32 neurons vs NGU very high
How about Jinc SR4 vs NGU Very High for Chroma?
AngelGraves13 is offline   Reply With Quote
Old 10th December 2016, 06:34   #41326  |  Link
khanmein
Registered User
 
Join Date: Oct 2012
Posts: 118
Quote:
Originally Posted by huhn View Post
d3d9 present in advanced is still broken. and NGU performance is still unacceptable slow on polaris.

power management is a lot better as least for me that's it.

i understand something else under "very much better"...
a lot AMD users said don't have issue & try ask them thru;

https://forum.lowyat.net/topic/4100453

http://www.overclock.net/f/67/amd-ati

this is y i'm NV fan boy!
khanmein is offline   Reply With Quote
Old 10th December 2016, 06:45   #41327  |  Link
burfadel
Registered User
 
Join Date: Aug 2006
Posts: 2,229
Quote:
Originally Posted by khanmein View Post
a lot AMD users said don't have issue & try ask them thru;

https://forum.lowyat.net/topic/4100453

http://www.overclock.net/f/67/amd-ati

this is y i'm NV fan boy!
Recent NVidia cards have been having issues as well. At least most of the issues with AMD have been resolved in the latest driver. I guess if D3D 9 mode is causing issues just used D3D 11 mode .
burfadel is offline   Reply With Quote
Old 10th December 2016, 07:02   #41328  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by AngelGraves13 View Post
How about Jinc SR4 vs NGU Very High for Chroma?
Jinc SR4 vs NGU Very High
ryrynz is offline   Reply With Quote
Old 10th December 2016, 07:13   #41329  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by StinDaWg View Post
chroma is way overrated
Quote:
Originally Posted by StinDaWg View Post
You will not convince me to waste gpu and use anything other than Bicubic 60AR
Quote:
Originally Posted by StinDaWg View Post
I see zero difference in real world viewing.
Let's try this using a similar scene then but larger, from the film Kubo and the Two Strings.

Bicubic 60 AR vs NGU Very High.

Last edited by ryrynz; 10th December 2016 at 07:17.
ryrynz is offline   Reply With Quote
Old 10th December 2016, 07:51   #41330  |  Link
AngelGraves13
Registered User
 
Join Date: Dec 2010
Posts: 255
Quote:
Originally Posted by ryrynz View Post
NGU looks pretty nice for Chroma...

One last request...

NGU Very High vs NGU Very High SR4

I want to see if SuperRes helps at all with Chroma for NGU
AngelGraves13 is offline   Reply With Quote
Old 10th December 2016, 08:09   #41331  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by AngelGraves13 View Post
I want to see if SuperRes helps at all with Chroma for NGU
It certainly makes things brighter. Is it better? Who knows, up to you to decide.

NGU Very High vs NGU Very High SR4
ryrynz is offline   Reply With Quote
Old 10th December 2016, 08:22   #41332  |  Link
CruNcher
Registered User
 
CruNcher's Avatar
 
Join Date: Apr 2002
Location: Germany
Posts: 4,926
Quote:
Originally Posted by ryrynz View Post
Let's try this using a similar scene then but larger, from the film Kubo and the Two Strings.

Bicubic 60 AR vs NGU Very High.

Nice example i guess everyone can percept the shaprness difference here like in Neo-XP his examples with NGU-Med

Your example becomes even more fascinating because of the lower filesize of the better perceptable psy result.

I mean we talking in your example about almost 10kb less for this particular frame with a overall better psy result.

But this was allready visible in Neo-XP his examples though not really with a final filesize reduction but the opposite.

So your example is kinda also crazy in it's overall result there seems to be a compression benefit as well now ?.

We definitely need some better more scientific analysis of what madshi does here
__________________
all my compares are riddles so please try to decipher them yourselves :)

It is about Time

Join the Revolution NOW before it is to Late !

http://forum.doom9.org/showthread.php?t=168004

Last edited by CruNcher; 10th December 2016 at 09:22.
CruNcher is offline   Reply With Quote
Old 10th December 2016, 08:37   #41333  |  Link
AngelGraves13
Registered User
 
Join Date: Dec 2010
Posts: 255
Quote:
Originally Posted by ryrynz View Post
It certainly makes things brighter. Is it better? Who knows, up to you to decide.

NGU Very High vs NGU Very High SR4
I see an improvement. The colors pop out more and the borders are thicker.
AngelGraves13 is offline   Reply With Quote
Old 10th December 2016, 13:03   #41334  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
more red on black comparison.

bicubic 60 AR: https://abload.de/img/bicubic60arq0ltz.png
bilateral sharp:https://abload.de/img/bilateralsharpuza5g.png
NGU low: https://abload.de/img/ngulowwbzq1.png
NGU very high: https://abload.de/img/nguveryhighqglpe.png
recon sharp: https://abload.de/img/reconstructionsharprjbmr.png
super XBR 100 AR: https://abload.de/img/superxbr100ar43a0c.png
super XBR 150 AR: https://abload.de/img/superxbr150arkra6a.png

super XBR 150 AR is doing an fantastic job for it's price on this image and bilateral is doing pretty bad here and that is the most obvious thing to "learn" here.

@StinDaWg

i'm pretty sure he got your point.
but the chroma scaler aren't there to waste electricity. people created these scaler and put a lot of work in them and they are not joking around.

if you don't see the difference FINE! but why do you have to stop them comparing chroma scaler?

@khanmein
nothing changed for madVR usages so whatever:
http://bugs.madshi.net/view.php?id=431
huhn is offline   Reply With Quote
Old 10th December 2016, 13:43   #41335  |  Link
StinDaWg
Registered User
 
Join Date: Jan 2014
Posts: 216
NGU doubling is really impressive with 720p animation @ 1080p. After playing back an episode of Family Guy in 720p and then 1080p version, it was hard to tell the difference. It's easy to tell the difference with Jinc and NNEDI3, 720p version looks blurry. Switch to NGU and bam, back in focus. The only thing that stands out is aliasing on some diagonal lines with NGU, otherwise they look almost identical.

Quote:
Originally Posted by huhn View Post
if you don't see the difference FINE! but why do you have to stop them comparing chroma scaler?
I'm not stopping anyone from doing anything. The differences to me are not noticeable in 99% of real world circumstances. All these comparisons of red text are not applicable to what you see watching in the majority of a video. I still haven't come across anything where I paused a video, changed chroma, then saw any difference at all. With cartoons, movies, sports or anything else.

But, if the only way to tell the difference is with these extreme fringe examples, then more power to them. That will be my last comment on the matter. It really doesn't matter to me what other people use. I'm just glad something like madVR exists for us videophiles.

Last edited by StinDaWg; 10th December 2016 at 13:54.
StinDaWg is offline   Reply With Quote
Old 10th December 2016, 13:46   #41336  |  Link
Neo-XP
Registered User
 
Neo-XP's Avatar
 
Join Date: Mar 2016
Location: Switzerland
Posts: 140
Quote:
Originally Posted by huhn View Post
more red on black comparison.

bicubic 60 AR: https://abload.de/img/bicubic60arq0ltz.png
bilateral sharp:https://abload.de/img/bilateralsharpuza5g.png
NGU low: https://abload.de/img/ngulowwbzq1.png
NGU very high: https://abload.de/img/nguveryhighqglpe.png
recon sharp: https://abload.de/img/reconstructionsharprjbmr.png
super XBR 100 AR: https://abload.de/img/superxbr100ar43a0c.png
super XBR 150 AR: https://abload.de/img/superxbr150arkra6a.png

super XBR 150 AR is doing an fantastic job for it's price on this image and bilateral is doing pretty bad here and that is the most obvious thing to "learn" here.

@StinDaWg

i'm pretty sure he got your point.
but the chroma scaler aren't there to waste electricity. people created these scaler and put a lot of work in them and they are not joking around.

if you don't see the difference FINE! but why do you have to stop them comparing chroma scaler?

@khanmein
nothing changed for madVR usages so whatever:
http://bugs.madshi.net/view.php?id=431
Yes, be careful with Bilateral, it can give you the best, but also the worst results depending on the situation.

Super-xbr is much more polyvalent, with values of 100-150.
I use NGU-Med at the moment (or NGU-High on new versions), because it is overall sharper and cleaner... and it is cold now where I live

Example with BilateralSharp / Bicubic60 AR / super-xbr150 / NGU-Low / NGU-Med / NGU-High :



I zoom for you in the right spot :



Now I let you decide which one to use.

Last edited by Neo-XP; 10th December 2016 at 15:04.
Neo-XP is offline   Reply With Quote
Old 10th December 2016, 14:04   #41337  |  Link
StinDaWg
Registered User
 
Join Date: Jan 2014
Posts: 216
Quote:
Originally Posted by huhn View Post
more red on black comparison.

bicubic 60 AR: https://abload.de/img/bicubic60arq0ltz.png
bilateral sharp:https://abload.de/img/bilateralsharpuza5g.png
NGU low: https://abload.de/img/ngulowwbzq1.png
NGU very high: https://abload.de/img/nguveryhighqglpe.png
recon sharp: https://abload.de/img/reconstructionsharprjbmr.png
super XBR 100 AR: https://abload.de/img/superxbr100ar43a0c.png
super XBR 150 AR: https://abload.de/img/superxbr150arkra6a.png

super XBR 150 AR is doing an fantastic job for it's price on this image and bilateral is doing pretty bad here and that is the most obvious thing to "learn" here.
I don't see any difference between these images. What am I supposed to be looking for? I guess it's possible that either my plasma or Firefox is doing something that is negating the effects, because they all look the same on this tv.
StinDaWg is offline   Reply With Quote
Old 10th December 2016, 14:15   #41338  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
try to compare bilateral and super XBR 150.

most obviously the red triangles but not limited to them.

i don't think it is possible to hide the differences with a sub sampling screen. maybe vivid mode or something like that can hide this but...

i'm pretty sure firefox can't hide this what so ever.

this is not a chroma heavy image but it still shows a huge difference with bilateral.
huhn is offline   Reply With Quote
Old 10th December 2016, 14:32   #41339  |  Link
StinDaWg
Registered User
 
Join Date: Jan 2014
Posts: 216
OK. I see a slight difference in sharpness/clarity with the red triangles between bilateral and super XBR 150. Less so with bicubic 60AR. If that's all I'm missing by using a faster algorithm, I'm fine with that. There's no way I could see that in motion anyways.
StinDaWg is offline   Reply With Quote
Old 10th December 2016, 14:37   #41340  |  Link
Zaoshi
Registered User
 
Join Date: Dec 2016
Posts: 14
Hello, I seem to be having some weird performance behaviour with NGU.

In one case madVR v0.91.3 NGU very high was taking 90 ms, but if I switch to NNEDI > Nearest > NGU very high it goes down to 16 ms for a dozen seconds and then back to 90 ms.: https://www.youtube.com/watch?v=Vt8FdGAgXL0

On madVR v0.91.4 similar situation has occurred on a different video. Using NGU high I watched few episodes with 30 ms / frame performance, but then it suddenly went up to 175 ms / frame. I tried previous trick and it went down to 40 ms / frame. While I was writing this post it has managed to go back to 30 ms / frame.: https://www.youtube.com/watch?v=tRxcFWl4qfA
Edit: it refuses to play at 30 ms / frame anymore. Had to revert back to NNEDI to continue watching.
Edit2: one episode later NGU works properly again.
Edit3: I think I figured it out. My GPU goes to sleep for some reason and reduces performance to save power and lower temperature. Any idea how to prevent this?

Not sure whether this is madVR issue or something with my system, but this never happened back when I was using NNEDI.

I have GeForce GTX 1070 with driver version 376.19.

Last edited by Zaoshi; 10th December 2016 at 22:06.
Zaoshi 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 20:34.


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