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 25th March 2016, 07:29   #37141  |  Link
Mano
Registered User
 
Join Date: Jul 2008
Posts: 54
just a note if someone is wondering 16.3 crimson amd creates random flicker horizon lines (the kind like in the horror movie effect) while playing video with madvr.
Mano is offline   Reply With Quote
Old 25th March 2016, 09:24   #37142  |  Link
QBhd
QB the Slayer
 
QBhd's Avatar
 
Join Date: Feb 2011
Location: Toronto
Posts: 697
Quote:
Originally Posted by Mano View Post
just a note if someone is wondering 16.3 crimson amd creates random flicker horizon lines (the kind like in the horror movie effect) while playing video with madvr.
Indeed... I noticed this, and took some screens shots. For me it was only when crispen edges was enabled. However, when I looked at the screen shots the effect was not there. I never thought any further of it, but yes... I guess it would be a driver related issue.

I know of one scene in The Big Short where it happens, if I have time I will track down which driver introduced this.

QB
__________________
QBhd is offline   Reply With Quote
Old 25th March 2016, 09:42   #37143  |  Link
Georgel
Visual Novel Dev.
 
Georgel's Avatar
 
Join Date: Nov 2015
Location: Bucharest
Posts: 200
Quote:
Originally Posted by ryrynz View Post
While it affects lines well for anime it's an edge thinner so it works on a lot of background elements you wouldn't want sharpened. I notice it especially on already thin things like bridge ropes etc.
This is why I have it working in tandem with Awarpsharp2 and line darkening in Avisynth, high level line thinning in anime can look pretty unnatural (super sharp & thin on some things and not on others)
Until we get line thinning/darkening it's the best we've got in madVR, just look at more than just the thick lines before you settle on a edge thinning value.
Edge thinning works very good with high values of edge sharpening.

Edge thinning alone would make things too thin.

I like the effects of ET 8 + ES 4 + AB150 + AR on both background and foreground elements.

Though too much edge thinning might change the meaning of some scenes, it is worth it. Especially where are lots of thick black lines. But it improves more!
Georgel is offline   Reply With Quote
Old 25th March 2016, 09:51   #37144  |  Link
Wolfy59
Registered User
 
Join Date: Oct 2011
Posts: 54
Hi and thanks for the work you do here. Does someone has an AMD R9 380 nitro oc 4 go and have lot of lags and drop frames when super res x2 anti bloating and AR are activated ?

Last edited by Wolfy59; 25th March 2016 at 10:06.
Wolfy59 is offline   Reply With Quote
Old 25th March 2016, 10:06   #37145  |  Link
Georgel
Visual Novel Dev.
 
Georgel's Avatar
 
Join Date: Nov 2015
Location: Bucharest
Posts: 200
How do we use profiles?

I am still having a hard time understanding where to start.
Georgel is offline   Reply With Quote
Old 25th March 2016, 12:41   #37146  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by Georgel View Post
How do we use profiles?
Good place to start.

https://imouto.my/tutorials/madvr/
ryrynz is offline   Reply With Quote
Old 25th March 2016, 12:45   #37147  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
Quote:
Originally Posted by Georgel View Post
Edge thinning works very good with high values of edge sharpening.

Edge thinning alone would make things too thin.

I like the effects of ET 8 + ES 4 + AB150 + AR on both background and foreground elements.

Though too much edge thinning might change the meaning of some scenes, it is worth it. Especially where are lots of thick black lines. But it improves more!
it is deforming, aliasing and it removes details.

http://screenshotcomparison.com/comparison/166725

it can be way way more harmful then this BTW.

but as you pleases.
huhn is offline   Reply With Quote
Old 25th March 2016, 13:05   #37148  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by huhn View Post
it is deforming, aliasing and it removes details.
http://screenshotcomparison.com/comparison/166725
it can be way way more harmful then this BTW.
Nice example, that's pretty bad. Are you using 8.0 after upscaling is complete? I find line thinning should only ever be used with that option. After every 2x upscale deforms lines too much.

I find even 4.0 is actually too high for anime but it works beautifully on just thick lines though, but too many unwanted side effects, hence why I keep it at ~2.0 + Awarpsharp.

I certainly would not use high values on 1080 content. Works quite well on 480-720 though.

Last edited by ryrynz; 25th March 2016 at 13:08.
ryrynz is offline   Reply With Quote
Old 25th March 2016, 13:35   #37149  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
it's not scaled at all.

even 2.0 with a 2x scaling is starting to deform.
of cause it looks kind of perfect compared to 8.0.

i personally wouldn't go over 0.5 on FHD and 1.0 on UHD(ok UHD is hard to judge).

but like always i don't like to change the picture. so nothing new.
huhn is offline   Reply With Quote
Old 25th March 2016, 14:05   #37150  |  Link
Georgel
Visual Novel Dev.
 
Georgel's Avatar
 
Join Date: Nov 2015
Location: Bucharest
Posts: 200
Quote:
Originally Posted by huhn View Post
it is deforming, aliasing and it removes details.

http://screenshotcomparison.com/comparison/166725

it can be way way more harmful then this BTW.

but as you pleases.
Why does the example deform so much? It is way different from all results I am getting.

I think that it is better for many anime out there.

Why did it induce aliasing for you? On my setup it eliminates any aliasing that was there before. I am confused.

There are very few anime that are 1080 natural, most are lower resolution, even if they are encoded to 1080.
Georgel is offline   Reply With Quote
Old 25th March 2016, 14:13   #37151  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
destroy it your self: http://abload.de/img/00004.m2ts_snapshot_0wju9u.png

sharp edges and line thinning are for sure not hiding aliasing.
huhn is offline   Reply With Quote
Old 25th March 2016, 14:17   #37152  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by leeperry View Post
OK my bad, it's just that Shiandow said "Improve guess by: Softening the image, Subtracting differences with the original image, Sharpening" and it sure sounds/looks like unsharp masking to me
I think that may be a description of Shiandow's original algorithm, when he introduced SuperRes. But a lot has changed since then.

Quote:
Originally Posted by YxP View Post
Could you elaborate on your personal feelings what SR actually does? In a way I have always thought about it as a sort final picture enhancer that I can count on to make the picture to be closer to the so called ground truth, even if I may not see the effects every frame of every possible video. Of course I can see, especially on animated content, if it brings unwanted artifacts, but maybe on a sort of theoretical level? For example, do you use it always at some strength? And yes, I have read the technical explanation.

[edit]

My point here being, what are the enhancements SR really requires for it to be better?
Don't really know what to say.

Quote:
Originally Posted by RainyDog View Post
One thing I've noticed is that anti-bloating has a positive effect on reigning in the brightness boosting of adaptive sharpen. On a black level setting test pattern, adaptive sharpen by itself reveals IRE 16 on a properly calibrated display. But with anti-bloating enabled the image is slightly darker so only 17 and above flash which is how it should be of course.

Adaptive sharpen plus anti-bloating is looking like it might be a formidable combination actually. AS maxed out at 3.0 combined with 150% AB might even be viable you know... Though more testing is probably needed to confirm that before we get too carried away
Glad you like it!

Quote:
Originally Posted by ryrynz View Post
I think you're getting mixed up or something is wrong, because AS5.png (AS 0.5) is clearly sharper than AS15_AB100.png
I guess it depends on how you define "sharpness". AS 0.5 has more "contrast", but less "focused lines". This is not a very scientific description, but maybe you understand what I mean.

My point in our discussion over the last 3 posts or so is that I consider the sharpening strength and the anti-bloating strength to be independent settings. You can think of the anti-bloating strength to be a slider which changes the sharpening "look". Basically at 0% anti-bloating you get 100% AdaptiveSharpen. If you go to 100% anti-bloating, you get something which looks more like crispen edges. If you use 50% anti-bloating, you get something in between AdaptiveSharpen and crispen edges, so to say.

So if you decrease both sharpening strength and anti-bloating strength, you're not achieving a similar end result. Decreasing both sharpening strength and anti-bloating strength at the same time basically changes both the look and the strength of the sharpening (so you get less strong but more bloated sharpening).

Quote:
Originally Posted by ryrynz View Post
Regarding anti-bloating with SSIM 2D @ 100% I think I'd prefer to keep it disabled. I could understand people perhaps wanting to enable 25 or 50% though.. beyond that you're getting similar results to just running SSIM at lower values.
I've tested with some more photos and with many of them it looks like your description fits well enough for SSIM downscaling. However, there are some cases where SSIM 100% with 100% anti-bloating looks very different from SSIM 25% (or turning SSIM off completely) without anti-bloating. I'll post screenshots later.

Quote:
Originally Posted by fedpul View Post
I found the problem, big queues again. Reset settings to default and then I start changing settings one to one and the crash was caused by the queues being set to max. It works well at default settings. Sorry for the report.
Ah, good to know!

I guess at some point I should probably check if I'm running out of GPU RAM. <sigh> So many things to do...

Quote:
Originally Posted by xul View Post
Yes, sorry madshi, in my case I think it was a "false alarm" as I was experimenting with various settings (LAV and madvr) to find which is the best for my configuration (Q9650@3,6ghz gtx970 MPC-HC x64) and I guess that I had NNEDI3 img doubing / supersampling which killed the performance..thats it.. now I'm using these settings and everything is smooth with great picture quality and no dropped frames during playback (using mostly 1080p contents on a 1920x1200 res monitor)...
Good to hear you got it solved.

Quote:
Originally Posted by JarrettH View Post
Nope, not either. SuperRes is different in that it digs into the picture and brings out real detail. I can't speak for the other refinements, but what SuperRes does puts it in a class of its own.
Ok, thanks for your feedback.

Quote:
Originally Posted by Warner306 View Post
I quickly repeated the anti-bloating test with SuperRes.

This time the images were scaled as follows:

480p -> 1080p (2.25x)

super-xbr100 + SuperRes (3) (no AR)

OR

super-xbr100 + SuperRes (3) + anti-bloating 100% (no AR)


The sharpening is less harsh than crispen edges + SuperRes. However, some areas of the image are still borderline and could use a little taming.

I think my mistake was not comparing the original SD images with the 1080p upscale, because I think the one upscaled with anti-bloating is closer to the original in visible detail. Maybe it is a little softer, but it seems more natural. This finding is the same for both tests.

So I would support the idea of anti-bloating being available with SuperRes. Perhaps you could make it a checkbox like anti-ringing and find a value that best suits each SuperRes level (1-4). SuperRes (1) likely wouldn't need it all; SuperRes (4) might need a value of at least 100%.
Thanks for the comparison. Unfortunately the photobucket server seemed too busy to make the downloads available properly, so I couldn't really look at the images in full size.
madshi is offline   Reply With Quote
Old 25th March 2016, 14:43   #37153  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
madVR v0.90.17 released

http://madshi.net/madVR.zip

Code:
* added a couple of image doubling super-xbr anti-bloating options
* fixed: "relaxed" vs "strict" option wasn't properly hidden for upscaling
I tried making the anti-bloating filter work for upscaling. For most algos I failed. However, I found a way to make it work for super-xbr image doubling.

Keep the feedback coming!
madshi is offline   Reply With Quote
Old 25th March 2016, 14:45   #37154  |  Link
fedpul
Registered User
 
Join Date: Feb 2014
Posts: 94
Hi madshi, I was watching a tv show last night and I got a black screen with 90.16 even with default queues, so I remember that a user said a few posts ago that D3D11 presentation was broken on nvidia and I think it is true (maybe it is a driver problem, I really don't know if it is realted to madVR). I changed to D3D9 presentation and watch 2 more episodes after that and everything was ok.

I would also like to share a nice 4k image to test downscaling:

https://imgur.com/6EVXKDB

It is a China building with detailed lines everywhere, and the fine details int he roof are perfect for testing downscaling.
fedpul is offline   Reply With Quote
Old 25th March 2016, 15:00   #37155  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
Quote:
Originally Posted by madshi View Post
I tried making the anti-bloating filter work for upscaling. For most algos I failed. However, I found a way to make it work for super-xbr image doubling.

Keep the feedback coming!
i wanted to ask about that because super xBR 150 looks bloated and my be useful with anti bloat.

but i thought upscaling and sharpening is totally different.

super xBR 150 is very cheap and effective but it has some serious issue.

but not working great i would say edit: nope i'm totally wrong here.

i'm using FHD to UHD:

edit:i had crisp edges at 4.0 disabling it fixed it.
i played to much around sorry. still a strange artifact.

source: http://abload.de/img/00004.m2ts_snapshot_0wju9u.png

the images are to big so i have to use a file host.
http://www.file-upload.net/download-...perxBR.7z.html

it is double lining at the leg.

edit2:
it looks like it is anti aliasing but takes some power/color from lines.
the anti aliasing effect is very welcome and the power/color is not that strong.
so i think it is an improvement. super xBR anti bloating: 75 looks most promising.

edit: fixed some things... i'm out for today. i'm just confusing my self and the rest.

Last edited by huhn; 25th March 2016 at 15:21.
huhn is offline   Reply With Quote
Old 25th March 2016, 15:12   #37156  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by fedpul View Post
I would also like to share a nice 4k image to test downscaling:

https://imgur.com/6EVXKDB

It is a China building with detailed lines everywhere, and the fine details int he roof are perfect for testing downscaling.
It's not bad, but a really good test image needs a combination of sharp edges, detailed textures and some nature (trees, bushes, flowers). The sharp edges in your image are great, but the texture detail is rather soft, and there's no nature. Which means that the image only tests some aspects of downscaling, but not all.

Quote:
Originally Posted by huhn View Post
super xBR is very cheap and effective but it has some serious issue.

but not working great i would say i'm using FHD to UHD:

edit:i had crisp edges at 4.0 disabling it fixed it.
i played to much around sorry.
So after dialing down crispen edges, how do you like it? And which serious issues does super-xbr have in your opinion?
madshi is offline   Reply With Quote
Old 25th March 2016, 15:18   #37157  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
super xBR 150 has seroius issue not super xBR in general.
some bright colors around dark lines for example.
there is no anti bloat 150 anyway.
i edit the post.

super xBR 75 anti bloat looks very promising! i was using 75 normal before.
huhn is offline   Reply With Quote
Old 25th March 2016, 15:20   #37158  |  Link
fedpul
Registered User
 
Join Date: Feb 2014
Posts: 94
Quote:
It's not bad, but a really good test image needs a combination of sharp edges, detailed textures and some nature (trees, bushes, flowers). The sharp edges in your image are great, but the texture detail is rather soft, and there's no nature. Which means that the image only tests some aspects of downscaling, but not all.
Thanks for the info, I would like if you or anyone can give me some good images to test scaling algos. BTW the new Super-XBR with anti-bloating set to 25 is awesome, I found it very similar to NNEDI 32 neurons, but without the performance hit.
fedpul is offline   Reply With Quote
Old 25th March 2016, 15:35   #37159  |  Link
fedpul
Registered User
 
Join Date: Feb 2014
Posts: 94
Is there any guide on how to take screenshots for making comparisons, I tried using the save image option from MPC-BE but the final image is really small if you take it from a dvd resolution movie for example. Thanks in advance.
fedpul is offline   Reply With Quote
Old 25th March 2016, 15:40   #37160  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
Quote:
Originally Posted by fedpul View Post
Is there any guide on how to take screenshots for making comparisons, I tried using the save image option from MPC-BE but the final image is really small if you take it from a dvd resolution movie for example. Thanks in advance.
disable FSE and or overlay mode and use alt +print to capture the screen.
if you do this in full screen the full screen picture is save in your clip board.
now open a picture program like paint/paint.net and press control+v. you can now save the image.

Last edited by huhn; 25th March 2016 at 16:08. Reason: you can know same the image....
huhn 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 14:19.


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