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 6th February 2017, 02:27   #42301  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
on the mario picture. yes it doesn't look good.
huhn is offline   Reply With Quote
Old 6th February 2017, 03:32   #42302  |  Link
Georgel
Visual Novel Dev.
 
Georgel's Avatar
 
Join Date: Nov 2015
Location: Bucharest
Posts: 200
In my video examples, the higher the algorithm is set, the better it looks.

I would probably not use medium either - which is puzzling since on the Mario image I wasn't bothered at all by it.
Georgel is offline   Reply With Quote
Old 6th February 2017, 07:15   #42303  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by madshi View Post
So why remove "low"? Is it really *that* terrible? If I compare it to NNEDI-16, IMHO NGU pixart "low" competes just fine. Or try comparing "low" to the regular (non-pixart) NGU variant!
Actually when I first tested this I don't think that NGU pix low actually activated and I was looking at Lanczos. I noticed this same thing occurring when I played with it on my GTX 960.
NGU pix low isn't terrible, but NNEDI3 16 handles thick low res lines more elegantly.

So I had a look vs NNEDI3 on the HTPC, the high preset is the closest in terms of visual appearance to NNEDI3 with how it enhances lines. Very high is more aggressive with parts of the image which is thinks are not lines and sometimes removes them (yes they're blurry but NNEDI3 does not do this) So in this way it seems to operate very much like NGU. Basically NGU just comes off as a tad aggressive with how it operates on lines to me, even with this version of it.

When I compare the original image upscaled with Lanczos to NNEDI3 64 neurons + my Avisynth profile it looks similar in appearance just more refined, lines are similar in shape and appearance. With NGU pix in places there's shrinking of lines, some areas it works nicely with line edges curving them naturally and in other places it just looks a bit off. NNEDI3 is more natural with lines and in places it's sharper too which just gives it a more drawn feel to it where as NGU just looks rather bland almost artifical in appearance by comparison, this is more important to me than anything else.

NGU pix might be a better fit for me with upscaling 720p content due to the extra resolution (will have to take a look) but for DVD level stuff I'm sticking with NNEDI3. Basically you'd have to improve on what I almost consider to be basically perfect for my use case. I guess I'll have to upload pictures to show you, my guess it will be hard to make me replace NNEDI3, it's a great alternative only IMO because obviously in certain situations it shines, just not for me where I need it most.

Last edited by ryrynz; 6th February 2017 at 10:30.
ryrynz is offline   Reply With Quote
Old 6th February 2017, 07:25   #42304  |  Link
Arm3nian
Registered User
 
Join Date: Jul 2014
Location: Las Vegas
Posts: 177
Quote:
Originally Posted by madshi View Post
1) Do you think NNEDI3 looks better? Or the new NGU pixart variant?
3) Can I replace NNEDI3 with NGU pixart in the next official build (pretty please)?
4) Can I replace super-xbr (for luma doubling, only) with NGU pixart in the next official build?
1) Even on low, I think NGU pixart does a better job of making a crap source look better, compared to max nnedi3 settings. It looks the cleanest, which nnedi3 was the king of.

3) Yes because NGU pixart looks better for low quality sources and NGU already looks better for high quality sources.

4) I've liked super-xbr because it looks good and requires little power to run. But NGU pixart on low runs just as fast and looks better than super-xbr for low quality sources. Regular NGU on low looks pretty much the same as super-xbr for high quality sources, and runs at the same speed. So I don't see the need for super-xbr anymore.
Arm3nian is offline   Reply With Quote
Old 6th February 2017, 08:14   #42305  |  Link
mogli
Registered User
 
Join Date: May 2015
Posts: 106
Quote:
Originally Posted by arm3nian View Post
4) i've liked super-xbr because it looks good and requires little power to run. But ngu pixart on low runs just as fast and looks better than super-xbr for low quality sources. Regular ngu on low looks pretty much the same as super-xbr for high quality sources, and runs at the same speed. So i don't see the need for super-xbr anymore.
This could (almost) be used as an argument to stay with super-xbr. I guess you talk about luma only so what chroma are you using?
mogli is offline   Reply With Quote
Old 6th February 2017, 09:09   #42306  |  Link
Arm3nian
Registered User
 
Join Date: Jul 2014
Location: Las Vegas
Posts: 177
Quote:
Originally Posted by mogli View Post
This could (almost) be used as an argument to stay with super-xbr.
How so?
Quote:
Originally Posted by mogli View Post
I guess you talk about luma only so what chroma are you using?
I don't see any differences between the chroma upscaling options, from different passes to different algorithms, on low resolution low quality video, so I used the same one.

For chroma doubling, you are restricted to choosing what you want based on the luma option, so I left that on default as well.

The changes are obvious in luma doubling. In the mario picture, pixart low doesn't look that great, but neither does nnedi3 compared to pixart very high. In my sources, with actual people, pixart low performs better than in the mario. I would choose it over any configuration with nnedi3 or superxbr. Again though, with my sources, veryhigh pixart looks very good, just like in the mario.
Arm3nian is offline   Reply With Quote
Old 6th February 2017, 10:22   #42307  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by Thunderbolt8 View Post
yeah, but since I hardly do any scaling atm, I am more interested just in changes of chroma upsampling and wondered if there are any differences between the regular algo and pixelart as well when it comes just to chroma upsampling
It's a different algorithm, so of course there are differences.

Quote:
Originally Posted by cyberscott View Post
I'd say the picture quality with NGU pixart Very High looks a little better, or equal, depending on the source material. Definitely not worse with any material. Shaving off a few milliseconds is always a nice bonus!

There is slight loss of sharpness to my eyes with NGU pixart set to Medium. What really surprised me is there isn't a huge drop in quality between Very High to Medium. It also saves an additional 5 to 7 milliseconds of render time at Medium which is impressive. If I had less capable hardware, I'd have no issue with using the Medium setting. I'll stick with Very High for now with SD, because I can.
Sounds good!

Quote:
Originally Posted by Hyllian View Post
Holy! I like "NGU pixart" with games!
Thank you - coming from you this means high praise! I'll reply to your question via PM.

Quote:
Originally Posted by Georgel View Post
I only use NNEDI3 at 16 neurons - NGU very high is slower than NNEDI at 16 neurons in same conditions.

Example :

NNEDI3 at 16 neurons is slower than NGU at high

NGU pixart at very high takes double the time to process compared to NNEDI3 at 16 neurons.

More precise examples:

NGU high ~5ms

NNEDI3 at 16 neurons ~8ms

NGU very high ~16ms

If everything else was kept the same.
Yes, but how does quality compare? FWIW, NGU pixart "high" currently doesn't exist. If you activate it, it uses NGU pixart "medium" for luma, but it uses a higher quality chroma setting compared to NGU pixart "medium". What would be interesting to me is if you prefer NNEDI3-16 or NGU pixart medium (in terms of image quality), when testing with real world sources?

Quote:
Originally Posted by ryrynz View Post
Actually when I first tested this I don't think that NGU pix low actually activated and I was looking at Lanczos. I noticed this same thing occurring when I played with it on my GTX 960.
NGU pix low isn't terrible, but NNEDI3 16 handles thick low res lines more elegantly.
I'm glad to hear that NGU pix low isn't terrible. It certainly wasn't terrible in my own tests, so I was a bit puzzled when you said it was. I'm not sure why it didn't activate in your case. If you can reproduce that somehow reliably, maybe you can find out the circumstances in which it doesn't activate? Does the OSD then still claim it's active?

Quote:
Originally Posted by ryrynz View Post
So I had a look vs NNEDI3 on the HTPC, the high preset is the closest in terms of visual appearance to NNEDI3 with how it enhances lines. Very high is more aggressive with parts of the image which is thinks are not lines and sometimes removes them (yes they're blurry but NNEDI3 does not do this) So in this way it seems to operate very much like NGU. Basically NGU just comes off as a tad aggressive to me even with this version of it.

When I compare the original image upscaled with Lanczos to NNEDI3 64 neurons + my Avisynth profile it looks similar in appearance just more refined, lines are similar in shape and appearance only slimmer. With NGU pix in places there's shrinking of lines, some areas it works nicely with line edges curving them naturally and in other places it just looks a bit off. NNEDI3 is more natural with lines and that's more important to me than anything else. NGU pix might be a better option for upscaling 720p content due to the extra detail there (will have to take a look) but for DVD level stuff I'm sticking with NNEDI3. Basically you'd have to improve on what I almost consider to be basically perfect for my use case. I guess I'll have to upload pictures to show you, my guess it will be hard to make me replace NNEDI3, it's a great alternative only IMO because obviously in certain situations it shines, just not for me where I need it most.
Interesting. Yes, screenshots would be helpful (please always include the unscaled source image, too).

FWIW, it seems most other users actually prefer NGU pixart over NNEDI3, or at least don't prefer NNEDI3, so I might end up removing NNEDI3 soon, unless I find evidence that NGU pixart is really clearly worse in some situations.

Quote:
Originally Posted by Arm3nian View Post
1) Even on low, I think NGU pixart does a better job of making a crap source look better, compared to max nnedi3 settings. It looks the cleanest, which nnedi3 was the king of.

3) Yes because NGU pixart looks better for low quality sources and NGU already looks better for high quality sources.

4) I've liked super-xbr because it looks good and requires little power to run. But NGU pixart on low runs just as fast and looks better than super-xbr for low quality sources. Regular NGU on low looks pretty much the same as super-xbr for high quality sources, and runs at the same speed. So I don't see the need for super-xbr anymore.

[...]

In the mario picture, pixart low doesn't look that great, but neither does nnedi3 compared to pixart very high. In my sources, with actual people, pixart low performs better than in the mario. I would choose it over any configuration with nnedi3 or superxbr. Again though, with my sources, veryhigh pixart looks very good, just like in the mario.
I'm happy to hear that - thanks!
madshi is offline   Reply With Quote
Old 6th February 2017, 10:55   #42308  |  Link
mogli
Registered User
 
Join Date: May 2015
Posts: 106
Quote:
Originally Posted by Arm3nian View Post
How so?
Well, you say they are pretty much the same except that NGU smooth looks better for low quality images. But then again speed does depend on chroma so super-xbr is still faster in some cases. Since I would prefer one algorithm that handles more cases over several different ones, overall I still prefer super-xbr for now.
mogli is offline   Reply With Quote
Old 6th February 2017, 11:08   #42309  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by madshi View Post
FWIW, it seems most other users actually prefer NGU pixart over NNEDI3, or at least don't prefer NNEDI3, so I might end up removing NNEDI3 soon, unless I find evidence that NGU pixart is really clearly worse in some situations.
They're probably not looking as closely as I am at the same sort of content, I'm picky about how natural upscaling and processing must look and NGU just appears a tad too artificial compared to NNEDI3 with lines.
I'll be able to show you this, but you may not deem the differences significant enough which has me a little concerned. Adding blue edges and a touch of sharpness seems to bring things closer in line to NNEDI3 but there's still that slightly more aggressive line creation/deletion going on which I don't like. Why does NGU pix very high look a good step more unnatural than high? Scaling up NNEDI3 looks the same regardless, the lines just get better though beyond 64 neurons almost seems pointless in many situations but it least it looks the same, with NGU you get one step forward and one step sideways, they're all quite different in appearance.

I guess I'm going to have to take a lot of screenshots..

Any way.. I can duplicate an issue with scaling although this applies to NNEDI3 on screen not NGU. If I start a video using NGU pix (let madVR decide everything) and then switch to NNEDI3 (any neurons, let madVR decide everything) NNEDI3 doesn't activate and I get whatever madVR chose (in this case bicubic 60 all the way). If I change neuron count then NNEDI3 applies properly. This is using D3D9 on the Intel HD4000.

Last edited by ryrynz; 6th February 2017 at 11:17.
ryrynz is offline   Reply With Quote
Old 6th February 2017, 11:31   #42310  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by ryrynz View Post
Why does NGU pix very high look a good step more unnatural than high?
There is no real "high" atm (see release notes). "high" is the same as "medium", but with better chroma doubling.

Unnatural or not, I'll wait for your screenshots.

Quote:
Originally Posted by ryrynz View Post
Any way.. I can duplicate an issue with scaling although this applies to NNEDI3 on screen not NGU. If I start a video using NGU pix (let madVR decide everything) and then switch to NNEDI3 (any neurons, let madVR decide everything) NNEDI3 doesn't activate and I get whatever madVR chose (in this case bicubic 60 all the way). If I change neuron count then NNEDI3 applies properly. This is using D3D9 on the Intel HD4000.
OpenCL initialization takes a while. So when activating NNEDI3 for the first time, madVR starts with a simple linear interpolation (so the image appears without delay), while OpenCL initialization is done in a background thread. Once OpenCL is ready to go, the image should automatically be re-rendered with NNEDI3.
madshi is offline   Reply With Quote
Old 6th February 2017, 11:39   #42311  |  Link
Arm3nian
Registered User
 
Join Date: Jul 2014
Location: Las Vegas
Posts: 177
Quote:
Originally Posted by mogli View Post
Well, you say they are pretty much the same except that NGU smooth looks better for low quality images. But then again speed does depend on chroma so super-xbr is still faster in some cases. Since I would prefer one algorithm that handles more cases over several different ones, overall I still prefer super-xbr for now.
I said NGU low looks and performs similarly to super-xbr. You can still choose higher settings for NGU if your hardware allows, which can produce a better image.

Same for pixart, to a greater extent probably, higher NGU levels look quite a bit better than low/super-xbr.

As for having one algorithm that works for all cases, you still have to make profiles for different content if you want optimal settings, so it doesn't really matter.
Arm3nian is offline   Reply With Quote
Old 6th February 2017, 11:39   #42312  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by madshi View Post
There is no real "high" atm (see release notes). "high" is the same as "medium", but with better chroma doubling.
Yup, am aware.

Quote:
Originally Posted by madshi View Post
OpenCL initialization takes a while..
This is what I'm seeing, thanks.
ryrynz is offline   Reply With Quote
Old 6th February 2017, 13:53   #42313  |  Link
YxP
Registered User
 
Join Date: Oct 2012
Posts: 99
Pixart doesn't seem to do 4X upscaling yet so I can only assume quality will be even better later. Tested it quickly watching older Simpsons and Futurama (meaning: closer to that hand-drawn style) and it works very, very well. Definitely will use it instead of NNEDI myself.
YxP is offline   Reply With Quote
Old 6th February 2017, 14:23   #42314  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by YxP View Post
Pixart doesn't seem to do 4X upscaling yet so I can only assume quality will be even better later. Tested it quickly watching older Simpsons and Futurama (meaning: closer to that hand-drawn style) and it works very, very well. Definitely will use it instead of NNEDI myself.
Sounds good! I'm not sure yet if I can make 4x pixart, we'll see. So you seem to prefer pixart quality over NNEDI3? Could you maybe post a comparison screenshot or two?
madshi is offline   Reply With Quote
Old 6th February 2017, 15:07   #42315  |  Link
YxP
Registered User
 
Join Date: Oct 2012
Posts: 99
Quote:
Originally Posted by madshi View Post
Sounds good! I'm not sure yet if I can make 4x pixart, we'll see. So you seem to prefer pixart quality over NNEDI3? Could you maybe post a comparison screenshot or two?
I guess I could, but I'm not sure if it's worth it since I can only use NNEDI16 before my GPU fan goes nuts. I can use NGU or pixart medium with almost completely silent fan, so it's more of a performance thing for me, although I DO like NGU more also quality-wise. I'll check later if I can find some good examples anyways.
YxP is offline   Reply With Quote
Old 6th February 2017, 18:17   #42316  |  Link
Damien147
Registered User
 
Join Date: Mar 2011
Posts: 380
I am happy that I chose rx 470 over rx 460
Still can't use NGU Very High but by seeing some people mentioning ''unnatural'' look I am calm with NGU med.
@Madshi confess it,you get paid by AMD and NVidiots....
I see in the future GPUs with ''MadVR optimized'' sticker in the box.
Damien147 is offline   Reply With Quote
Old 6th February 2017, 18:30   #42317  |  Link
pose
Registered User
 
Join Date: Jul 2014
Posts: 69
Quote:
Originally Posted by Damien147 View Post
Still can't use NGU Very High
Why? What resolution youre scaling to?
pose is offline   Reply With Quote
Old 6th February 2017, 18:48   #42318  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
because NGU is very slow on polaris GPU for still unknown reasons.
huhn is offline   Reply With Quote
Old 6th February 2017, 18:53   #42319  |  Link
cork_OS
Registered User
 
cork_OS's Avatar
 
Join Date: Mar 2016
Posts: 160
Quote:
Originally Posted by ryrynz View Post
With NGU pix in places there's shrinking of lines, some areas it works nicely with line edges curving them naturally and in other places it just looks a bit off. NNEDI3 is more natural with lines and in places it's sharper too which just gives it a more drawn feel to it where as NGU just looks rather bland almost artifical in appearance by comparison, this is more important to me than anything else.
Seems I see what you mean (EDIT: NNEDI is sharper and lines are darker or more expressed/emphasized): http://screenshotcomparison.com/comparison/199707
Source sample (frame 80).
__________________
I'm infected with poor sources.

Last edited by cork_OS; 6th February 2017 at 19:11.
cork_OS is offline   Reply With Quote
Old 6th February 2017, 19:07   #42320  |  Link
pose
Registered User
 
Join Date: Jul 2014
Posts: 69
Quote:
Originally Posted by huhn View Post
because NGU is very slow on polaris GPU for still unknown reasons.
I have RX 470 and NGU very high is as fast as NNEDI3 128 in the same scenario. Probably not as fast as it should be but good enough for me.

Last edited by pose; 6th February 2017 at 19:11.
pose 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 15:56.


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