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 4th November 2015, 17:12   #34041  |  Link
Luv
Registered User
 
Join Date: Nov 2009
Posts: 63
Quote:
Originally Posted by XRyche View Post
My MSI R7 265 is not bad either especially for it's price point. It's like having a R9 270. I know it is by no means top of the line but it does a decent job with madVR and NNEDI3 Image doubling.
HD7790 (= + or - 260x),with Omega driver:
- D3D11
- chroma up:super-xbr (100)
- image down: Spline (4)
- image doubling: none
- image upscaling: Jinc
- dithering: ED 2
- No quality trade

The results are plain smashing (Even with interlaced samples).Big thanks,Madshi!
For me,89.12 is almost perfect (Almost because the codec in use isn't indicated in the OSD anymore and what does "Touch window from inside" mean?).
__________________
7/32,ZP Max 10,LAV 0.66,madVR 89.12,HD 7790,Omega driver 14.12.
Luv is offline   Reply With Quote
Old 4th November 2015, 17:26   #34042  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
Quote:
Originally Posted by AngelGraves13 View Post
I prefer Jinc AR for Chroma, and Image Upscaling. SuperXBR can be too sharp and I don't have enough time in my life to test out the optimal sharpness.
super-xbr has very big problems with ringing, you have to sacrifice a lot of sharpness to get to an acceptable ringing level if the source isn't optimal.

Quote:
Originally Posted by AngelGraves13 View Post
SupeRes is pointless as well, as it likely just sharpens the image and I literally can't tell what it does after a good hour of viewing with/without it.
madshi gave us some more SuperRes options with the latest builds, you could try a sharpness of 3 (and maybe 2 passes instead + linear light).
It should be very visible in most cases, but of course additional sharpness gets harder to notice with increasing viewing distance.
aufkrawall is offline   Reply With Quote
Old 4th November 2015, 18:02   #34043  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
Quote:
Originally Posted by Luv View Post
For me,89.12 is almost perfect (Almost because the codec in use isn't indicated in the OSD anymore
works fine with h264, vc-1 and mpeg2 like it always did.
Quote:
and what does "Touch window from inside" mean?).
that's how scaling is done it kind of means scale the image but don't crop it.
huhn is offline   Reply With Quote
Old 4th November 2015, 21:37   #34044  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Quote:
Originally Posted by aufkrawall View Post
super-xbr has very big problems with ringing, you have to sacrifice a lot of sharpness to get to an acceptable ringing level if the source isn't optimal.


madshi gave us some more SuperRes options with the latest builds, you could try a sharpness of 3 (and maybe 2 passes instead + linear light).
It should be very visible in most cases, but of course additional sharpness gets harder to notice with increasing viewing distance.
super-xbr has about as much ringing as Jinc in the test images I've seen. According to the chart in madVR, it is one of the best algorithms with regards to ringing. It is not that bad at all. I use it all the time.

These comparison shots show super-xbr is superior:

Image Comparison – Clown:
Jinc
super-xbr100
NNEDI3 256 neurons

Image Comparison – Lighthouse:
Jinc
super-xbr100
NNEDI3 256 neurons

Image Comparison – Lighthouse Top:
Jinc
super-xbr100
NNEDI3 256 neurons

Last edited by Warner306; 4th November 2015 at 21:51.
Warner306 is offline   Reply With Quote
Old 4th November 2015, 22:29   #34045  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
With the lighthouse top example, you can see that with a sharpness of 100, the ringing with super-xbr is clearly more distinct than with Jinc AR.
And the top needle (and dark line structures in general) gets very fat, which is another disadvantage.
aufkrawall is offline   Reply With Quote
Old 4th November 2015, 22:32   #34046  |  Link
Luv
Registered User
 
Join Date: Nov 2009
Posts: 63
Quote:
Originally Posted by huhn View Post
works fine with h264, vc-1 and mpeg2 like it always did.


that's how scaling is done it kind of means scale the image but don't crop it.
Thanks for the explanations,Huhn.I just rebooted and everything is back to normal.
__________________
7/32,ZP Max 10,LAV 0.66,madVR 89.12,HD 7790,Omega driver 14.12.
Luv is offline   Reply With Quote
Old 4th November 2015, 22:36   #34047  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Quote:
Originally Posted by aufkrawall View Post
With the lighthouse top example, you can see that with a sharpness of 100, the ringing with super-xbr is clearly more distinct than with Jinc AR.
And the top needle (and dark line structures in general) gets very fat, which is another disadvantage.
Sure the ringing is more distinct, but not larger. This is due to the detail brought-out by super-xbr, which gives it an overall advantage. With real world content, super-xbr is not nearly as distracting as Bicubic100 or Lanczos4, with regards to ringing. It is actually pretty good at avoiding excessive ringing.

I have heard a couple of people say super-xbr rings too much. But I don't think this is true.

Last edited by Warner306; 4th November 2015 at 22:39.
Warner306 is offline   Reply With Quote
Old 4th November 2015, 23:12   #34048  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
Quote:
Originally Posted by Warner306 View Post
Sure the ringing is more distinct, but not larger. This is due to the detail brought-out by super-xbr, which gives it an overall advantage.
I'm afraid it's not just details what super-xbr brings out.
More intensive haloing and very fat lines are not to be found in the source.

If you watch cartoons with ringing right along black contoure lines, it is very annoying (and lines also get too fat).
It gets even uglier if you put SuperRes on top and super-xbr 100 tends also more to aliasing than NNEDI3 64.

With bacondither's new Adaptive Sharpen experimental build (not yet included in madVR), you can get more sharpness with NNEDI3 as well, without ugly haloing. The same goes for NNEDI3 + SuperRes.

Of course Jinc isn't magic, yes. It's a limited linear scaler. But with super-xbr, you get a very soft image if you don't want ringing or dark lines boosted.
Probably less of an issue for many BDs, but there's always a first time when some artifacts become annoying.
aufkrawall is offline   Reply With Quote
Old 4th November 2015, 23:51   #34049  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Quote:
Originally Posted by aufkrawall View Post
I'm afraid it's not just details what super-xbr brings out.
More intensive haloing and very fat lines are not to be found in the source.

If you watch cartoons with ringing right along black contoure lines, it is very annoying (and lines also get too fat).
It gets even uglier if you put SuperRes on top and super-xbr 100 tends also more to aliasing than NNEDI3 64.

With bacondither's new Adaptive Sharpen experimental build (not yet included in madVR), you can get more sharpness with NNEDI3 as well, without ugly haloing. The same goes for NNEDI3 + SuperRes.

Of course Jinc isn't magic, yes. It's a limited linear scaler. But with super-xbr, you get a very soft image if you don't want ringing or dark lines boosted.
Probably less of an issue for many BDs, but there's always a first time when some artifacts become annoying.
I've read madshi prefers super-xbr to Jinc. And past posts have found NNEDI3 and super-xbr are the best algorithms to use with SuperRes because they are sharper. SuperRes + Jinc leads to an image that appears bloated compared to SuperRes + super-xbr.

It is all subjective. The haloing is not noticeable with real-world content, to me, and the extra detail is very apparent.

I'm just standing up for super-xbr as a great algorithm as has been said about Jinc and NNEDI3. Other's mileage will vary.

This a quote from madshi:

Quote:
Originally Posted by madshi View Post
I've decided to make a big screenshot Based on that, here's another comparison, comparing different upscaling algorithms, followed by SuperRes:

Bilinear+SuperRes -|- Jinc+SuperRes -|- super-xbr+SuperRes -|- NNEDI3+SuperRes -|- GroundTruth

What we can see here is that SuperRes works well even when using Bilinear upscaling. However, SuperRes does *not* remove aliasing artifacts caused by the upscaling algorithm. E.g. look at the roof edges of the left two towers. Both Bilinear and Jinc have aliasing problems there. super-xbr and NNEDI3 have not. Because of this reason, my recommendation would be to use either super-xbr or NNEDI3, followed by SuperRes, for best image quality. The difference between super-xbr and NNEDI3 is pretty small, if you follow it up with SuperRes with high strength. So using super-xbr should save some precious GPU performance. Using Jinc+SuperRes might be an option, too, but you'll likely get more aliasing problems compared to super-xbr+SuperRes.

Last edited by Warner306; 4th November 2015 at 23:58.
Warner306 is offline   Reply With Quote
Old 5th November 2015, 00:48   #34050  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
Well, as I said, super-xbr may look mostly fine with most BD content, so it's absolutely legitimate to like it.

I'm aware that super-xbr reconstructs lines much better than Jinc with increasing scaling factor.
However, sharpness values over 75 can introduce a lot of artifacts. This doesn't have to be an issue for a specific picture, but there are definitely cases where it can look far from good.

For instance, you can hardly use SuperRes in linear light with super-xbr 100, as lines will get extremely fat. And you may not use gamma light either, since super-xbr 100 already noticeably highers many areas of the picture.
In fact, I even find super-xbr 100 alone with your Lighthouse example unconvenient to the eyes. It looks like a very artificial contrast to me. The line boosting of SuperRes LL is almost harmless, compared to super-xbr 100.

In the end, it's a matter of taste. However, I think it's a fact that super-xbr can't be combined well with most postprocessing, unlike NNEDI3.
aufkrawall is offline   Reply With Quote
Old 5th November 2015, 07:33   #34051  |  Link
JarrettH
Registered User
 
Join Date: Aug 2004
Location: Canada
Posts: 860
I can't see wanting to use additional post processing on top of a sharp scaler. A picture should look effortless.
JarrettH is offline   Reply With Quote
Old 5th November 2015, 11:52   #34052  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
Postprocessing isn't evil by definition. Hardly anyone would suggest to use Jinc or super-xbr without AR filter. In fact, afair it is always enabled for super-xbr in the image doubling settings, since it rings so terribly.
aufkrawall is offline   Reply With Quote
Old 5th November 2015, 21:39   #34053  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Quote:
Originally Posted by JarrettH View Post
I can't see wanting to use additional post processing on top of a sharp scaler. A picture should look effortless.
SuperRes softens super-xbr. So the picture is not oversharpened. It is actually improved.
Warner306 is offline   Reply With Quote
Old 5th November 2015, 22:18   #34054  |  Link
Thunderbolt8
Registered User
 
Join Date: Sep 2006
Posts: 2,197
Quote:
Originally Posted by Warner306 View Post
SuperRes softens super-xbr. So the picture is not oversharpened. It is actually improved.
but if its really improved from a rather objective point of view then it would be a default setting, wouldnt it?
__________________
Laptop Lenovo Legion 5 17IMH05: i5-10300H, 16 GB Ram, NVIDIA GTX 1650 Ti (+ Intel UHD 630), Windows 10 x64, madVR (x64), MPC-HC (x64), LAV Filter (x64), XySubfilter (x64) (K-lite codec pack)
Thunderbolt8 is offline   Reply With Quote
Old 5th November 2015, 22:34   #34055  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
As I already said, SuperRes doesn't like ringing.
I wouldn't use SuperRes with super-xbr. Instead I'd use super-xbr with a low sharpness setting + Adaptive Sharpen, then ringing is far less of a problem.

NNEDI3 64 quadrupling + SR (strength 3, sharpness 1, ll off):


super-xbr 100 + SR:


Jinc AR + SR:


super-xbr simply looks terrible in this example. I'd even prefer Jinc, as it has less ringing problems.
NNEDI3 + SuperRes is best by far.
aufkrawall is offline   Reply With Quote
Old 5th November 2015, 22:56   #34056  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Your findings back-up the previous test images. super-xbr has more ringing but less aliasing. That seems pretty clear. SuperRes was said to have the most problems with aliasing, which is why it was recommended to be used with NNEDI3 or super-xbr.

The simple nature of animation makes it unappealing as a test for image upscaling. Faces and detailed backgrounds with many surfaces show the differences in detail far better. But that is just my opinion.

I still like super-xbr. I don't think there is a right or wrong answer. The desire to have a sharp vs soft picture would have a lot to do with this preference.

Last edited by Warner306; 5th November 2015 at 23:11.
Warner306 is offline   Reply With Quote
Old 5th November 2015, 23:12   #34057  |  Link
har3inger
Registered User
 
Join Date: Feb 2014
Posts: 139
As far as I've seen, super-XBR doesn't really add much ringing from upscaling, but rather preserves source ringing a lot more than nnedi3 does. Also, in terms of sharpness, SuperXBR 75 is about the same as jinc3, so those two would be a fairer comparison for how badly each rings.

A bigger problem seems to be the way small details with high contrast (pupils in eyes) get fairly messed up when sharpness is set lower than 2 for superres. In the blonde's eyeballs, the pupils get extra white dots in them that shouldn't be there. If it's anything like the images I've been testing, the problem fixes itself at sharpness 2 and 3 (but 3 tends to be pretty aliased).

For clean sources, SuperXBR + SuperRes for 720p->1080p (CR downscale) has no ringing problems in my experience. I don't think it's fair to compare it to nnedi3, which is a lot better at cleaning up source ringing artifacts compared to any other madvr scaler.
har3inger is offline   Reply With Quote
Old 5th November 2015, 23:16   #34058  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Quote:
Originally Posted by har3inger View Post
As far as I've seen, super-XBR doesn't really add much ringing from upscaling, but rather preserves source ringing a lot more than nnedi3 does. Also, in terms of sharpness, SuperXBR 75 is about the same as jinc3, so those two would be a fairer comparison for how badly each rings.

A bigger problem seems to be the way small details with high contrast (pupils in eyes) get fairly messed up when sharpness is set lower than 2 for superres. In the blonde's eyeballs, the pupils get extra white dots in them that shouldn't be there. If it's anything like the images I've been testing, the problem fixes itself at sharpness 2 and 3 (but 3 tends to be pretty aliased).

For clean sources, SuperXBR + SuperRes for 720p->1080p (CR downscale) has no ringing problems in my experience. I don't think it's fair to compare it to nnedi3, which is a lot better at cleaning up source ringing artifacts compared to any other madvr scaler.
At 720p -> 1080p, I notice the image appears "fatter" when Jinc is used with SuperRes as opposed to super-xbr + SuperRes. I assuming this difference comes down to aliasing on edges.
Warner306 is offline   Reply With Quote
Old 5th November 2015, 23:25   #34059  |  Link
Ver Greeneyes
Registered User
 
Join Date: May 2012
Posts: 447
Quote:
Originally Posted by har3inger View Post
A bigger problem seems to be the way small details with high contrast (pupils in eyes) get fairly messed up when sharpness is set lower than 2 for superres. In the blonde's eyeballs, the pupils get extra white dots in them that shouldn't be there. If it's anything like the images I've been testing, the problem fixes itself at sharpness 2 and 3 (but 3 tends to be pretty aliased).
That's pretty interesting. I've been wondering what settings for the new SuperRes would be good - I figured Sharpness = 1 was just a reasonable default, but it sounds like that's not the case (I didn't try resetting my settings to find out what the defaults were).
Ver Greeneyes is offline   Reply With Quote
Old 5th November 2015, 23:42   #34060  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
Quote:
Originally Posted by Warner306 View Post
SuperRes was said to have the most problems with aliasing, which is why it was recommended to be used with NNEDI3 or super-xbr.
I suspect this is a huge misunderstanding in general regarding SuperRes.
SuperRes reintroduces aliasing of the source when it has been filtered away by upscaling (NNEDI3 has a very strong AA effect).
I haven't seen a single case where SuperRes really introduces aliasing which isn't supposed to exist.

There have been ground truth comparisons here which showed aliasing problems with SuperRes, yes. But the aliasing was probably introduced by downscaling the image with linear light first.
This is e.g. the case with the lighttower example of leeperry. When you downscale it in gamma light, there is no aliasing visible when upscaling it again and applying SuperRes.

We btw. also still have the radius option in madVR, a higher value will lead to less aliasing (but probably also less sharpness).

In fact, SuperRes can even repair aliasing introduced by upscaler. But this functionality is not endless, so an upscaling algorithm with fewest aliasing is still preferred (which is again NNEDI3 with enough neurons).

Quote:
Originally Posted by Warner306 View Post
The simple nature of animation makes it unappealing as a test for image upscaling. Faces and detailed backgrounds with many surfaces show the differences in detail far better. But that is just my opinion.
I have to disagree. (Mostly) clean lines show the characteristics of an algorithm very well, which makes it easy to spot weaknesses.
Of course this alone is not comprehensive enough to judge finally, but I still find it very important.


Quote:
Originally Posted by har3inger View Post
As far as I've seen, super-XBR doesn't really add much ringing from upscaling, but rather preserves source ringing a lot more than nnedi3 does.
I'd rather call it boost instead of preserve.
But on the other hand it's true that NNEDI3 filters away ringing, yes. SuperRes reintroduces some of it, since it's like correcting an upscaling error.

Quote:
Originally Posted by har3inger View Post
Also, in terms of sharpness, SuperXBR 75 is about the same as jinc3, so those two would be a fairer comparison for how badly each rings.
It was meant in context of SuperRes fixing upscaling errors.
I think it just doesn't work well with super-xbr, as even tiny amounts of ringing can get nasty with super-xbr + SR (even with a sharpness option of 1, default is 2 which boosts ringing a lot more).

Quote:
Originally Posted by har3inger View Post
A bigger problem seems to be the way small details with high contrast (pupils in eyes) get fairly messed up when sharpness is set lower than 2 for superres. In the blonde's eyeballs, the pupils get extra white dots in them that shouldn't be there. If it's anything like the images I've been testing, the problem fixes itself at sharpness 2 and 3 (but 3 tends to be pretty aliased).
I think it's a problem of the source, I wouldn't put too much meaning into it.
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 19:42.


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