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 19th December 2016, 14:57   #41601  |  Link
Neo-XP
Registered User
 
Neo-XP's Avatar
 
Join Date: Mar 2016
Location: Switzerland
Posts: 140
Quote:
Originally Posted by huhn View Post
ok got it amplifying again...

i mean a very very small part of an 720p image was scaled to 720p
The source is FHD (1080p). What you see in my comparison is a part of the resulting 4K image.
Neo-XP is offline   Reply With Quote
Old 19th December 2016, 15:18   #41602  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
this is a 720p part of the scaled UHD image in 720p from the "hat": https://abload.de/img/720puhdcutkuocr.png

these image from here are amplified: http://forum.doom9.org/showpost.php?...ostcount=41614

NGU very high/chroma NGU med: https://abload.de/img/bicubic60art1r0o.png
NGU very high/chroma bicubic 60 AR: https://abload.de/img/ngumedokrt7.png

i'm pretty sure lanczos will not make a difference comparing these two.
BTW you can compare lanczos 3 AR and bicubic 60 AR upscaling to get an idea of the difference in chroma which looks like nothing at all to me.
huhn is offline   Reply With Quote
Old 19th December 2016, 15:26   #41603  |  Link
Neo-XP
Registered User
 
Neo-XP's Avatar
 
Join Date: Mar 2016
Location: Switzerland
Posts: 140
Quote:
Originally Posted by huhn View Post
this is a 720p part of the scaled UHD image in 720p from the "hat": https://abload.de/img/720puhdcutkuocr.png

these image from here are amplified: http://forum.doom9.org/showpost.php?...ostcount=41614

NGU very high/chroma NGU med: https://abload.de/img/bicubic60art1r0o.png
NGU very high/chroma bicubic 60 AR: https://abload.de/img/ngumedokrt7.png
You didn't test with Lanczos there, so what is the point ?

Quote:
Originally Posted by huhn View Post
i'm pretty sure lanczos will not make a difference comparing these two.
Yes it makes a difference, I've posted the result and Lanczos performs better than Bicubic
It is not a big difference, but to quote madshi : "highest quality has priority over anything else".

Quote:
Originally Posted by huhn View Post
BTW you can compare lanczos 3 AR and bicubic 60 AR upscaling to get an idea of the difference in chroma which looks like nothing at all to me.
I have already done that, and Bicubic is not a good algorithm to upscale.
Can you show an image where bicubic 60 AR is better for upscaling than lanczos 3 AR ?
Neo-XP is offline   Reply With Quote
Old 19th December 2016, 15:49   #41604  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
you are showing a difference by amplifying an image to a ridiculous point.
if you need to amplify this than there is no difference it is that simple.

there is no real difference between NGU and bicubic except the arm where NGU is doing bad looks cut out and adds ringing. but NGU chroma doubling does change the image.

if you really want a lanczos 3 AR vs bicubic 60 AR i can do that...
here you go Cr rendered as Y:
https://abload.de/img/1dup4r.png
https://abload.de/img/24arvq.png
huhn is offline   Reply With Quote
Old 19th December 2016, 16:05   #41605  |  Link
Neo-XP
Registered User
 
Neo-XP's Avatar
 
Join Date: Mar 2016
Location: Switzerland
Posts: 140
Quote:
Originally Posted by huhn View Post
you are showing a difference by amplifying an image to a ridiculous point.
if you need to amplify this than there is no difference it is that simple.
Sorry, but this make no sense. If there was no difference, zooming on the images would produce the same result, which is not the case.

Quote:
Originally Posted by huhn View Post
there is no real difference between NGU and bicubic except the arm where NGU is doing bad looks cut out and adds ringing. but NGU chroma doubling does change the image.
I see a lot of differences between Bicubic and NGU for chroma doubling, even at 100% zoom.

Quote:
Originally Posted by huhn View Post
if you really want a lanczos 3 AR vs bicubic 60 AR i can do that...
here you go Cr rendered as Y:
https://abload.de/img/1dup4r.png
https://abload.de/img/24arvq.png

https://abload.de/img/24arvq.png is Lanczos3 AR, right ? because it looks sharper.
I don't know how you did that, but it is cool and you can see the differences better

I don't think you can show an image where bicubic 60 AR is better for upscaling than lanczos 3 AR.

Last edited by Neo-XP; 19th December 2016 at 16:23.
Neo-XP is offline   Reply With Quote
Old 19th December 2016, 16:24   #41606  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
Quote:
Sorry, but this make no sense. If there was no difference, zooming on the images would produce the same result, which is not the case.
so you are now going to watch stuff with a zoom?
you can show a difference always so we should get scaler in 100 steps with enough zooming it will show a difference.

see doesn't make sense.

Quote:
I see a lot of differences between Bicubic and NGU for chroma doubling, even at 100% zoom.
this image is not zoom and i can see the difference too. NGU and bicubic are totally different scaler while lanczos and cubic scalers are similar.
Quote:
https://abload.de/img/24arvq.png is Lanczos3 AR right ? because it looks sharper.

Can you show an image where bicubic 60 AR is better for upscaling than lanczos 3 AR ?
to be honest i don't even know but i guess it is and it doesn't matter showing Cr only is a very hard amplifying.
the difference is close to nothing it is a waste of time.
on a normal image the Y channel will totally dominate the image.

high quality chroma scaling is questionable small most of the time but the difference are day night if Cr is rendered as Y.

and about "highest quality has priority over anything else".
chroma doubling was remove from madVR before and the main reason/image example it is in again was a 320p pixel art RGB image that showed a difference (it wasn't really better BTW.)

not to talk about even higher bit deep that could be used and other stuff that may show a difference when we zoom in enough.
huhn is offline   Reply With Quote
Old 19th December 2016, 17:01   #41607  |  Link
Neo-XP
Registered User
 
Neo-XP's Avatar
 
Join Date: Mar 2016
Location: Switzerland
Posts: 140
Quote:
Originally Posted by huhn View Post
the difference is close to nothing it is a waste of time.
Yes, I know, but why propose an inferior algorithm to upscale when you could have Lanczos3 AR with the same performance ?

Of course you won't be able to differentiate them in motion at 100% zoom, but that is not the point.
The point, for me at least, is to have the highest quality you can get for your hardware

Bicubic shouldn't be used to upscale anything at all. It is good for downscaling though (with AR).

Last edited by Neo-XP; 19th December 2016 at 18:02.
Neo-XP is offline   Reply With Quote
Old 19th December 2016, 17:10   #41608  |  Link
Mistery 73
Registered User
 
Join Date: Oct 2016
Posts: 31
Quote:
Originally Posted by cork_OS View Post
"Reduce ringing artifacts" madVR filter only removes haloing, not mosquito noise. So you shold try deringing & mosquito noise reduction Avisynth filters.
I am also interested in reducing mosquito noise as I have to use the plagin? I have to enter in the madVR folder?
thank you
Mistery 73 is offline   Reply With Quote
Old 19th December 2016, 18:52   #41609  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
I've come across a user trying to create profiles for 3D and non-3D content. This is his rule:

else if (3D) and ((srcWidth > 1280) and (srcWidth <= 1920)) "1080p 3D"
else if (3D) and ((srcWidth <= 1280) and (srcHeight > 720) and (srcHeight <= 1080)) "1080p 3D"

else if (not 3D) and ((srcWidth > 1280) and (srcWidth <= 1920)) "1080p"
else if (not 3D) and ((srcWidth <= 1280) and (srcHeight > 720) and (srcHeight <= 1080)) "1080p"

Can you tell me if anything is wrong with this rule?
Warner306 is offline   Reply With Quote
Old 19th December 2016, 19:07   #41610  |  Link
cork_OS
Registered User
 
cork_OS's Avatar
 
Join Date: Mar 2016
Posts: 160
Quote:
Originally Posted by Mistery 73 View Post
I am also interested in reducing mosquito noise as I have to use the plagin? I have to enter in the madVR folder?
thank you
I was not expecting I will explain the Avisynth usage basics on this forum.
To use Avisynth plugins you need to install one of numerous Avisynth builds, find and download Avisynth plugin on this forum or avisynth.nl, then manually write Avisynth script on its scripting language and open script in video player. You should expect compatibility, stability and speed issues between particular Avisynth build and particular build of particular Avisynth plugin.
__________________
I'm infected with poor sources.

Last edited by cork_OS; 19th December 2016 at 19:09.
cork_OS is offline   Reply With Quote
Old 19th December 2016, 20:08   #41611  |  Link
cork_OS
Registered User
 
cork_OS's Avatar
 
Join Date: Mar 2016
Posts: 160
Sorry, previous post turned up too sarcastic and uninformative.
madVR include no avisynth / vapoursynth support, so opening video in player through avisynth script is the only way of real-time madVR preprocessing.
In addition to overall avisynth usage complexities, a lot of avisynth plugins are still too slow for real-time usage. So madVR existence is the salvation for end users and I pray for it's further development.
Lack of bitrate (and compression artifacts) is the usual thing for nowadays dowloadable / streaming H.264 content, as well as compression artifacts on older MPEG-2 DVDs (and countless MPEG-4 ASP encodes), and I do not know how to reduce compression artifacts while preserving details in real time.
__________________
I'm infected with poor sources.
cork_OS is offline   Reply With Quote
Old 19th December 2016, 21:33   #41612  |  Link
Georgel
Visual Novel Dev.
 
Georgel's Avatar
 
Join Date: Nov 2015
Location: Bucharest
Posts: 200
Quote:
Originally Posted by hannes69 View Post
Yeah, it also contributes to climate change using GPUs like many here which are drawing more power alone than my whole desktop computer system including the monitor... Maybe power is way too cheap in many countries...
It is definitely a question of efficiency. Is it efficient to invest 500% more power to gain 0.5% better picture quality
The trend here is obvious: Spending as much money as one wants for a certain GPU and then maxing the load (90% until frames are dropping). And if max drawing algorithms arentīt possible, throw away the half year old GPU and buying a new one. Reminds me sometimes of the gaming scene, there obviously it seems very important to get 500 frames per second for certain games, for whatever thatīs good for.
I think for the people who are so proud of their GPU handling 256 neuron NNEDI3 luma and chroma quadrupling thereīs some need for the possibility of 512 neuron octopling
No offense, only a way of thinking about that.
But that's the whole fun!

Seriously now, I find myself using Jinc more than any other algo, except maybe for NNEDI since those two are best for my usage scenario, but NNEDI 200% forced makes my laptop go too loud so I can't really enjoy the whole anime or movie or AMV I'm watching... Guess Jinc will do for now.
Georgel is offline   Reply With Quote
Old 19th December 2016, 21:43   #41613  |  Link
Georgel
Visual Novel Dev.
 
Georgel's Avatar
 
Join Date: Nov 2015
Location: Bucharest
Posts: 200
Quote:
Originally Posted by Neo-XP View Post
This is what I use now after a lot of testing with films (on good old v0.91.1 with a GTX 970) :

artifact removal : reduce banding artifacts (medium + high)

chroma upscaling : NGU-Med / super-xbr150 + AR + SR4 (no clear winner here, I use NGU-Med because the edges are sharper)
image downscaling : Bicubic150 + AR (relaxed)
image doubling for luma : 2x supersampling with NGU-Med
image doubling for chroma : Lanczos3 + AR
upscaling refinement : soften edges 1 + add grain 3

dithering : Ordered + use colored noise + change dither for every frame
Error Diffusion - option 2 gives better result, but I can't use it for 4K with my GC (90% load and dropped frames).

To remove ringing and dark halos, I just use this in aviSynth :
SetMemoryMax(768)
SetMTMode(3)
ffdshow_source()
SetMTMode(2)
FineDehalo(thmi=128, thma=128, thlimi=50, thlima=50, contra=1.0)
SetMTMode(1)
GetMTMode(false) > 0 ? distributor() : last

I'm still waiting for a new version to come out to disable image quadrupling and use Lanczos3 + AR for chroma doubling (NGU is overkill for this and Bicubic60 is not as sharp). I will then be able to do an apples-to-apples comparison.

Speaking of comparisons, here is one between super-xbr150 + AR + SR4 and NGU-Med for chroma upscaling :



Edges are sharper with NGU, but the texture looks sharper with super-xbr... I can't decide.
With NGU, if the source is too low quality, that sharpness does not look natural, but it works well for RL videos compared to drawn materials.

Super XBR is also a bit sharp for edges, depending on the material.
Georgel is offline   Reply With Quote
Old 20th December 2016, 01:32   #41614  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
When recommending settings, how much performance does Error Diffusion take versus Ordered dithering at 4K? At 1080p, the performance is negligible with a 750 Ti. I wonder if it saves enough performance to bother using Ordered.

How many ms does Ordered save you? For me, it's a couple ms at most.

Last edited by Warner306; 20th December 2016 at 01:35.
Warner306 is offline   Reply With Quote
Old 20th December 2016, 02:21   #41615  |  Link
Neo-XP
Registered User
 
Neo-XP's Avatar
 
Join Date: Mar 2016
Location: Switzerland
Posts: 140
Quote:
Originally Posted by Warner306 View Post
When recommending settings, how much performance does Error Diffusion take versus Ordered dithering at 4K? At 1080p, the performance is negligible with a 750 Ti. I wonder if it saves enough performance to bother using Ordered.

How many ms does Ordered save you? For me, it's a couple ms at most.
Typically, with Error Diffusion option 2 versus Ordered, 20% more load on a GTX 970 (from 58% to 78%) and +10ms (from 24ms to 34ms) for 1080p content upscaled to 4K.

I don't know why I had more than 90% load before, I can't reproduce the issue at the moment

You really have to test on specific patterns to see any difference between the two.

Last edited by Neo-XP; 20th December 2016 at 02:29.
Neo-XP is offline   Reply With Quote
Old 20th December 2016, 04:23   #41616  |  Link
MistahBonzai
Registered User
 
Join Date: Mar 2013
Posts: 101
Quote:
Originally Posted by Mistery 73 View Post
I am also interested in reducing mosquito noise as I have to use the plagin? I have to enter in the madVR folder?
thank you
You should be familiar with the ins/outs of AviSynth (MT and Plus32/64) before embarking on what may initially seem a simple task. Otherwise it may well turn into a rat-hole where problem solving rather than media consumption provides the entertainment/pleasure.
MistahBonzai is offline   Reply With Quote
Old 20th December 2016, 04:36   #41617  |  Link
oddball
Registered User
 
Join Date: Jan 2002
Posts: 1,264
I have been using MadVR to switch display modes. However. When switching to 2160p when the TV out is initially set to 1080 the video starts in a quarter size window in the top left corner of my LG OLED (LG TV reads 2160p on info). If the TV out is initially set to 2160p it plays UHD video fine but when I play a 1080p video it switches to 1080 on the TV and plays a portion of the video only on one half of the TV (again showing 1080 on TV info). The other side is black. If I disable DX11 in MadVR (as long as FSE is enabled) it works as expected, but DX9 is buggy as heck on AMD drivers. If I just set 2160p as the resolution to switch to MadVR will scale to 2160p which I presume puts more strain on the GPU? Which is the correct method? Should I forget about sending 1080p to my UHD TV and let MadVR scale it to 2160p?

I find the different resolutions confusing when it comes to using MadVR. It was pretty straightforward before
oddball is offline   Reply With Quote
Old 20th December 2016, 06:18   #41618  |  Link
EncodedMango
Registered User
 
Join Date: Jun 2013
Posts: 65
Is it normal to have such high rendering times for 10-bit videos? I've noticed it in both HEVC and H264. I suspect it might be because of a lack of H/W Decoding but I'd like to confirm regardless.


GTX 970, LAV with DXVA2 native is what I use by the way.

Last edited by EncodedMango; 20th December 2016 at 06:21.
EncodedMango is offline   Reply With Quote
Old 20th December 2016, 06:28   #41619  |  Link
AngelGraves13
Registered User
 
Join Date: Dec 2010
Posts: 254
Lower your settings. You're dropping way too many frames.
AngelGraves13 is offline   Reply With Quote
Old 20th December 2016, 06:42   #41620  |  Link
khanmein
Registered User
 
Join Date: Oct 2012
Posts: 118
@sidspyker disable smooth motion can at least reduce 10 ms
khanmein 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 13:52.


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