madVR - high quality video renderer (GPU assisted)

[huhn]

destroy it your self: http://abload.de/img/00004.m2ts_snapshot_0wju9u.png

sharp edges and line thinning are for sure not hiding aliasing.

[madshi]

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 :o
I think that may be a description of Shiandow's original algorithm, when he introduced SuperRes. But a lot has changed since then.

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.

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!

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).

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.

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...

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.

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.

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]

madVR v0.90.17 released

http://madshi.net/madVR.zip

* 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!

[fedpul]

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.

[huhn]

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-11423763/superxBR.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.

[madshi]

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.

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?

[huhn]

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.

[fedpul]

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 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.

[huhn]

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.

[fedpul]

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 know same the image.

Thanks huhn I will make some comparisons later.

[leeperry]

added a couple of image doubling super-xbr anti-bloating options
right on time, :thanks: for the new build.

anyway, just ran more tests today:
-AR is pointless for SR as it doesn't ring to begin with
-AB is also pointless for SR as feeding a blurry picture to SSIM looks ugly
-the best results for me are with both chroma & luma SR@1 + NNEDI3 doubling + 0.7 enhance detail + SSIM 2D 25% with 75% AB. It retains SSIM's attention to details without looking digital, yay!
-my only gripe is that this setup makes NNEDI3's supremacy very obvious as in comparison NEDI looks blurry and sxbr25 artificial, but maybe spicing up the latter with some AB would straighten things out somehow :devil:

[clsid]

I guess at some point I should probably check if I'm running out of GPU RAM. <sigh> So many things to do.Implementing this sooner than later may give a lot of return on investment. Certainly now that 4K is becoming more common. Dealing with OOM related 'bug' reports is a huge waste of your time.

Maybe the GUI could also show an indication of the (worst case) memory requirements for a given queue size. That should help users to avoid too large queues. Assume the largest of the available display devices. Assume smooth motion is on, and whatever else may increase memory usage. Show estimate for common input sizes.

GPU queue size: 8 (requires up to xxx MB for 1080p and xxxx MB for 4K)

[Neo-XP]

@madshi

Is it possible to include the parameters of FineSharp (sstr, cstr, lstr, pstr, ldmp, xstr and xrep) to madVR ?

If not in the GUI, maybe with a config file ?

[Anima123]

Super-xBR has been improved to another level, IMO.

Somehow, unlike some others in the thread, I prefer SSIM 1D than 2D for 576p -> doubler -> 1080p chain, 1D seems more neat for my taste.

Will upload comparison when I get home.

[aufkrawall]

As if it wasn't already confusing enough:
Even in FSE, I'm getting one dropped frame after 5 minutes of watching. So I disabled DXVA2 copyback in LAV to see if it makes any difference, and indeed: Now queues also get filled properly in windowed fullscreen and I can watch more than 5 minutes without any issues so far. However, still low level monitoring with driver must be used in Afterburner.
Hope they fix this at some point in the drivers with new GPUs (got the same problem with Sapphire Trixx OC tool instead of Afterburner) or else one might need to use an external GPU (e.g. of APU) for decoding if one wants to use voltage control or custom fan curve with HEVC 10 bit content.
Story's not over yet. :cool:
There are also problems with 1080p 60fps upscaling to 1440p fullscreen/borderless without Afterburner ever started when DXVA2 copyback is used (in borderless queues stop filling at some point and with FSE, queues get almost completely emptied once after exactly 5 minutes of playback).
It doesn't occur with 1080p 30fps. I have the vague feeling that something's simply broken or at least very shaky with the Hawaii decoder for 1080p 60fps in general. Maybe there's some weird overhead or stupid bug that can trigger this behavior, dunno.
With software playback, it's completely fine.
Well, that makes me a bit more hopeful regarding the situation with future GPUs, Hawaii always had some video decoding problems (e.g. always crashed with bsod when decoding 4k).

[videonerd]

Hello can someone please tell me if AR and/or LL are required for:

chroma upscaling: super-xbr100
image downscaling: bicubic150
image upscaling: jinc
upscaling refinement: superres3

Thank you!

[iSeries]

Hi, upscaling enhancements don't seem to work for me when supersampling 1080p > 1080p with the latest two versions - they don't have any effect.

[Warner306]

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.

If you use OneDrive in Windows 10, there is a setting to automatically save screenshots in the OneDrive -> Pictures folder. This saves the step of having to paste the image into an image editing program.

[Warner306]

Could someone post a screenshot comparison of super-xbr100 with anti-bloating enabled?

[Warner306]

Hi, upscaling enhancements don't seem to work for me when supersampling 1080p > 1080p with the latest two versions - they don't have any effect.

You need to use really high values. SuperSampling is not great for that reason alone. Image enhancements applied to a 1080p source will do more to improve the image.

[Warner306]

Hello can someone please tell me if AR and/or LL are required for:

chroma upscaling: super-xbr100
image downscaling: bicubic150
image upscaling: jinc
upscaling refinement: superres3

Thank you!

chroma upscaling: super-xbr100 (AR)
image downscaling: bicubic150 (AR + LL)
image upscaling: jinc (AR)
upscaling refinement: superres3 (no AR, maybe anti-bloating 50-75%)

[iSeries]

You need to use really high values. SuperSampling is not great for that reason alone. Image enhancements applied to a 1080p source will do more to improve the image.

I should say i have zoom small black bars away so a 1.85.1 film fills the screen (so not technically 2x supersampling but close). However the upscaling enhancements have zero effect, no matter what i set them to, and render times do not change. Reverting back to .15 and it works fine, render times go up and image changes.

[fedpul]

I have made an album with some comparisons of chroma and doubling algos. By now I have uploaded just 4 images and they are based on a DVD release of the movie Howl's Moving Castle.

The link of the album is: http://imgur.com/a/eRXn8

My intention was to compare the new SXBR with NNEDI3 which is considered the best algorithm.

Please feel free to comment!

[perlita.pajares]

I cannot deal with playing 3D bluray content. I have a i7 4700mq with Intel HD4600 and dgpu Nvidia GTX765M with optimus. When playing 3d I cannot use the nvidia GPU due to the HDMI port is attached to the HD4600 card.

I am using MPC-HC with LAV filters and MADVR, always the latest nightly builds of them. The problem is that even setting madvr to the lowest settings (nearest neighbor in chroma and the rest deactivated), the player (MPC-HC) unexpectly stop and back me to desktop after 30 sec of playing without more information. I have used DVXA copy-back, native and Intel Quicksync without a successful result. MADVR with and without FSE and with D3D11 and D3D9, all possible combinations.

Do you have any sugestion or solution?

PD: no problem with powerdvd.

Thank you so much in advance.

[AngelGraves13]

I should say i have zoom small black bars away so a 1.85.1 film fills the screen (so not technically 2x supersampling but close). However the upscaling enhancements have zero effect, no matter what i set them to, and render times do not change. Reverting back to .15 and it works fine, render times go up and image changes.

You probably have "disable scaling if image size changes by only" in Zoom Control.

[Georgel]

destroy it your self: http://abload.de/img/00004.m2ts_snapshot_0wju9u.png

sharp edges and line thinning are for sure not hiding aliasing.

I had tested again and again to be able to reproduce your results.

I found out how.

For me, ET + ES + AB150 + AR will eliminate aliasing that WAS there.

Here is the trick: If I am rendering to a 1080 screen, I must use either upscaling refinement if the video is under 1080 OR image enhancements if the video is 1080.

Upscaling refinement takes significantly longer to process (rendering time) because I have it set to apply after every 2X step, but produces perfect results.

Same can be said about image enhancements, if used only when video is already the same resolution as display.

At such high values, there will be *small* distortion, but it still better than without those algorithms.

There are anime that are already over-smooth, and those would be distorted. On this same note, I seriously think that the reason I like ET + ES + AR + AB at such high values is that I personally have either bad encodes, or anime that were drawn too fuzzy for these resolutions. I checked with some more, especially new anime, and it works magic for half, while it distorts the other half, inducing aliasing and distorting objects.

@madshi Is there any possibility of inducing an aggressive anti aliasing algorithm?

[chros]

here a sample with a 3D LUT.
http://www.file-upload.net/download-11420168/3DLUTslowdownissue.7z.html

i can't trigger this issue without using native UHD. zooming in at 1080p doesn't trigger it.
I couldn't see the difference either.

I cannot deal with playing 3D bluray content. I have a i7 4700mq with Intel HD4600 and dgpu Nvidia GTX765M with optimus. When playing 3d I cannot use the nvidia GPU due to the HDMI port is attached to the HD4600 card.
Is it also the case with 2D content? (I don't have a 3D capable display.)

was micro sttuter on some random scenes, some are reporterd presentation glitche s some not
like 7 presentation glitches on 50 minutes playing...
testing more i solve this checking the option use a separate device for presentantion... and mpc-hc internal audio renderer exclusive mode
now the movies are very smooth with some normal flickering
It didn't solve it on my system.
Madshi, have you seen my edited post regarding to this? http://forum.doom9.org/showthread.php?p=1761445#post1761445
Tell me if you need a debug log or anything else.

[Presco]

Hello
when you set a scaling level of personalized office 125% I can not access some options madVR.

Thank you for making the window of adjustable madVR.

http://nsm08.casimages.com/img/2016/03/26//1603261113066677614098184.jpg (http://www.casimages.com/img.php?i=1603261113066677614098184.jpg)

http://nsm08.casimages.com/img/2016/03/26//1603261111096677614098183.jpg (http://www.casimages.com/img.php?i=1603261111096677614098183.jpg)

[ryrynz]

Hello
when you set a scaling level of personalized office 125% I can not access some options madVR.


Already reported, the options will change in future and will be fixed in due course, consider current builds new features to be a WIP.

[huhn]

I had tested again and again to be able to reproduce your results.

I found out how.

For me, ET + ES + AB150 + AR will eliminate aliasing that WAS there.

Here is the trick: If I am rendering to a 1080 screen, I must use either upscaling refinement if the video is under 1080 OR image enhancements if the video is 1080.

Upscaling refinement takes significantly longer to process (rendering time) because I have it set to apply after every 2X step, but produces perfect results.

Same can be said about image enhancements, if used only when video is already the same resolution as display.

At such high values, there will be *small* distortion, but it still better than without those algorithms.

There are anime that are already over-smooth, and those would be distorted. On this same note, I seriously think that the reason I like ET + ES + AR + AB at such high values is that I personally have either bad encodes, or anime that were drawn too fuzzy for these resolutions. I checked with some more, especially new anime, and it works magic for half, while it distorts the other half, inducing aliasing and distorting objects.

@madshi Is there any possibility of inducing an aggressive anti aliasing algorithm?

and why don't you post a screen than?

[XMonarchY]

Wow! With earlier versions of madVR I was unable to force 2x SuperSampling for Luma Doubling NNEDI3 32n + Chroma Upscaling NNEDI3 32n in 1080p videos without severe frame drops, but now I can. However, it only works fine in Exclusive FullScreen mode. In windowed mode, I still get bad frame drops. Why is that? Maybe SuperSampling does not actually work in Exclusive FullScreen mode???

[huhn]

maybe start with an OSD screen?

[iSeries]

You probably have "disable scaling if image size changes by only" in Zoom Control.

That option doesn't seem to change the behaviour. If I untick 'automatically detect black bars', upscaling enhancements work fine with supersampling, clear difference to picture and render times shoot up to around 35ms. As soon as I tick 'automatically detect black bars' render times drop dramatically and the upscaling enhancements 'crispen edges' and 'superres' no longer have an effect, even though the image is still being doubled and then downscaled by setting '2x supersampling' in image doubling.

[huhn]

@iSeries

sound like a bug.

do you have steps to reproduce the issue maybe even from default settings?

i will check it out too.

you can backup your settings file if you don't want to lose your current settings.

[leeperry]

I think AB is too strong in the new sxbr options, I'm afraid an additional knob would be very much needed there :o

I got some 720p videos that give a dark pixelated look when NNEDI3 is used for luma only, I'll try to find what triggers it.

Somehow, unlike some others in the thread, I prefer SSIM 1D than 2D for 576p -> doubler -> 1080p chain, 1D seems more neat for my taste.

Fair enough, I guess we have a choice between the softer 1D with a higher strength and the sharper 2D with a lower one......at this point it boils down to the sharpness of the source material, how sharp your doubler is, how your display reacts and indeed personal preference. Bicubic also hides defects Jinc makes obvious.

But good point, I'd rather get 1D to work at it seems less demanding and as much as my 7850 can do all I want(FRC/SR/RS/SSIM/NNEDI3) on SD that's a no-go on 720p@1080p so I'm gonna have to either cut corners on doubling or downscaling.

[ashlar42]

madshi, always appreciate your continued development of madVR.

Considering the tons of options you now offer, I hope that sooner or later, when you reach the 1.0 milestone, you'll create complete and proper documentation for madVR. I know that, personally, that would make me extremely happy about buying a commercial release of it.

Thanks again. Keep up the good work.

[BluesFanUK]

Wow! With earlier versions of madVR I was unable to force 2x SuperSampling for Luma Doubling NNEDI3 32n + Chroma Upscaling NNEDI3 32n in 1080p videos without severe frame drops, but now I can. However, it only works fine in Exclusive FullScreen mode. In windowed mode, I still get bad frame drops. Why is that? Maybe SuperSampling does not actually work in Exclusive FullScreen mode???

I can't, most of my vids are shot in 1080p/60fps. If I even attempt x2 Supersamling I get horrendous frame drops, and I have a Palit 980ti Super Jetstream using just a 1080p monitor.

Same when I try to use Reconstruction for Chrome Upscaling. It takes my GPU usage right up to 100% and drops loads of frames to the point it bogs the PC down.

[perlita.pajares]

Is it also the case with 2D content? (I don't have a 3D capable display.)

No problem with 2D content, only with 3D.

Does someone have any idea?

[Dclose]

There's been a lack of love for Spline, so I thought I'd add some. At least for much of the stuff I watch, it works well, and doesn't use a lot of power.

It's like a slightly less sharp Lanczos or strong Bicubic, but with fewer artifacts and smoother macroblocking/noise. It uses less power than Jinc. I don't recall as I'm typing this, but I think it's sharper than Jinc. And much sharper than SoftCubic.

It uses noticeably less GPU than Jinc. madVR can do impressive things at full throttle, but it's good to also have settings that don't push the GPU and wattage and heat. Spline uses more power than certain settings, but barely more.

I've been going back and forth between Spline, Bilateral, and Reconstruction Soft for chroma. Bilateral and RS tend to win out, but they're all good.

For cranking up the GPU more but still not wanting to go all out, Super-XBR in the upscaling tabs and in image doubling seems to be great most of the time. Turning on image doubling even with Super-XBR basically doubles my GPU usage, but still uses less than power-hungry NNEID3.

Just adding some Spline love so it hopefully stays around.

[huhn]

i did some test for super xbr anti bloat

FHD -> UHD
source: http://abload.de/img/00004.m2ts_snapshot_0wju9u.png
spline3 AR: https://www.ezphotoshare.com/images/2016/03/26/TrvLI.png
super xbr 100 anti bloat: https://www.ezphotoshare.com/images/2016/03/26/TrXIA.png
super xbr 100: https://www.ezphotoshare.com/images/2016/03/26/Trl1D.png

HD -> QHD
source: https://www.ezphotoshare.com/images/2016/03/26/TrZLg.png
spline3 AR: https://www.ezphotoshare.com/images/2016/03/26/TrJlx.png
super xbr 100 anti bloat: https://www.ezphotoshare.com/images/2016/03/26/TrSpB.png
super xbr 100: https://www.ezphotoshare.com/images/2016/03/26/TrPdp.png

super xbr anti bloat is less sharp than spline 3 AR. which makes it really soft for upscaling.
the effect of anti bloat is looks useful. it has an very nice anti ringing effect. is it possible to use it with super xbr 125 or even 150 in the future?

[Anima123]

Where's the anti-bloating option for Super-xBR? I couldn't find one.

[huhn]

it's in the drop down menu.

http://abload.de/img/superxbr8ouen.png

[alf6]

But you do have the Windows 7 Platform Update installed? And you don't get a picture, neither in windowed nor in fullscreen exclusive mode?


Yeah I do. Anyways problem went away after a clean install of drivers, so probably not madVR related after all:)

Sorry for wasting your time, and thanks for the good work :)

[Warner306]

i did some test for super xbr anti bloat

FHD -> UHD
source: http://abload.de/img/00004.m2ts_snapshot_0wju9u.png
spline3 AR: https://www.ezphotoshare.com/images/2016/03/26/TrvLI.png
super xbr 100 anti bloat: https://www.ezphotoshare.com/images/2016/03/26/TrXIA.png
super xbr 100: https://www.ezphotoshare.com/images/2016/03/26/Trl1D.png

HD -> QHD
source: https://www.ezphotoshare.com/images/2016/03/26/TrZLg.png
spline3 AR: https://www.ezphotoshare.com/images/2016/03/26/TrJlx.png
super xbr 100 anti bloat: https://www.ezphotoshare.com/images/2016/03/26/TrSpB.png
super xbr 100: https://www.ezphotoshare.com/images/2016/03/26/TrPdp.png

super xbr anti bloat is less sharp than spline 3 AR. which makes it really soft for upscaling.
the effect of anti bloat is looks useful. it has an very nice anti ringing effect. is it possible to use it with super xbr 125 or even 150 in the future?

In those examples, I don't really like it. super-xbr should be sharp.

Does super-xbr still ring? If not, then the sharper version would be my preference. I am starting to feel the same way about SuperRes. Unless the bloating is obvious and offensive, anti-bloating will take away too much image detail that should be there. I have three image that show this pretty clearly. I'll post them some time in the next two days.

[huhn]

Does super-xbr still ring? If not, then the sharper version would be my preference. I am starting to feel the same way about SuperRes. Unless the bloating is obvious and offensive, anti-bloating will take away too much image detail that should be there. I have three image that show this pretty clearly. I'll post them some time in the next two days.

compared to spline 3 ar and super xbr 100, 100 anti bloat has less ringing. line look simply more natural but it is simply not sharp enough.

and 125 and 150 looks forced. and i see potential for anti bloat here. i hope it is at least as sharp as super xbr 100 but with the anti bloat effect.

[Warner306]

compared to spline 3 ar and super xbr 100, 100 anti bloat has less ringing. line look simply more natural but it is simply not sharp enough.

and 125 and 150 looks forced. and i see potential for anti bloat here. i hope it is at least as sharp as super xbr 100 but with the anti bloat effect.

Yes, I did notice anti-bloating tames ringing, so it may have its place. Using a higher value of super-xbr may be the ideal combination.

The only complaint I ever had about super-xbr was ringing. When combined with SuperRes, it looks so similar to NNEDI3 in test images. I would guess super-xbr is the most popular algorithm because of its price/performance ratio, so I would support improving it in any way.

[bcec]

In those examples, I don't really like it. super-xbr should be sharp.

Does super-xbr still ring? If not, then the sharper version would be my preference. I am starting to feel the same way about SuperRes. Unless the bloating is obvious and offensive, anti-bloating will take away too much image detail that should be there. I have three image that show this pretty clearly. I'll post them some time in the next two days.

I am with you. I don't like AB for super-xbr, it takes away all the sharpness and detail, and you end up with an artificially soft image (fhd->uhd). I don't find it useful for SuperRes either.

Initially I very much liked AS 1 (or 1.5) with AB 100% as an upscaling refinement, but after watching a full movie with it, things looked weirdly off. Too sharp, and too soft at the same time. Not sure how to explain. I am trying SharpenEdges 1.0 with AB 100% to see if that combination makes more sense.

[AngelGraves13]

My new settings are...

Chroma: Super-XBR 150 w/ AR + SR 1
Upscaling: Jinc AR
Downscaling: Jinc AR + AB 150
Image Refinement: SR 1 (every 2X) w/ AR + AB 150

I used Futurama for testing as it's easy to spot the differences in animated content. Lines are sharp and smooth. Thanks madshi!

Also, I find anything higher than 1 for Super Res is pointless and introduces ringing.

[Warner306]

I am with you. I don't like AB for super-xbr, it takes away all the sharpness and detail, and you end up with an artificially soft image (fhd->uhd). I don't find it useful for SuperRes either.

Initially I very much liked AS 1 (or 1.5) with AB 100% as an upscaling refinement, but after watching a full movie with it, things looked weirdly off. Too sharp, and too soft at the same time. Not sure how to explain. I am trying SharpenEdges 1.0 with AB 100% to see if that combination makes more sense.

I didn't watch an entire movie, but sharpen edges (0.5) with anti-bloating 100% looked like an improvement at 1080p -> 1080p. Removing fat lines is an obvious benefit. Taming oversharpened images also works well. But, it I get the sharpening at an acceptable level where it doesn't look oversharpened, it doesn't seem necessary.