[madshi]

Originally Posted by aufkrawall

I really like the new alternative color space option.
It gives an anti aliasing effect, but doesn't seem to introduce AA artifacts. However, strength should be highered in return to make contoures not look too thin.
I prefer algo 3: 1 changes brightness for some colors too much and 2 is too soft.
I still don't see a real gain in more passes than 1.
So my new favorite settings for SuperRes are:
passes 1, strength 0.7, algo 3, alternative color space used.
At least it looks great with the low res cartoon example.

Could you please try some other content, too? Preferable also some sharp and some soft real life video instead of Anime, maybe also with ringing and without ringing. These new SuperRes options may look different with different kinds of material. Of course I need to find values which work best as an allround solution.

Originally Posted by aufkrawall

Edit: madshi, would it be possible to use lower values than 0.1 with AdaptiveSharpen with the next version?

Ok, no big problem.

Originally Posted by tFWo

My experience with new SR: I turned it off as i struggle to find the purpose and actual refinement in this image refinement.

720p movie i tried actually looked softer and fuzzier with any of the SR algos. Alt. color space made it even worse. Turning SR (with alt color space on) off was like removing a thin layer of dirt

On SD movie (the one i used for those images in my last post) i can see the difference clearly between different algos when zoomed. 0 is sharpest, 2 is softest. 3 is different (sharper than 2). Combination of algo 2 + alt color space is nightmare fuel - Rings of Saturn .

Turning SR on with low number of passes with strength 1.00 doesn't do much but more importantly doesn't do anything positive to the picture. Increasing number of passes just brings more artifacts with negligible increase in sharpness.

What is the purpose of this new SR?

Rings of Saturn, hehe...

Originally Posted by Halfpant.exe

Liking the new SR changes
I watch a lot of low quality 480p videos
& passes 2, strength. 75 with algo 3 & alternative color space turned on is giving a fairly nice results
& for hd content i am using
algo 0 with alternative color space turned off & adaptive sharpen to 1.5

If you had to choose one setting for all sources, which would it be?

Originally Posted by leandronb

I edited my post with images, i think the file is 23p, but i think the resolution is a bit weird, but don't know. the osd is in the images.

Yes, the resolution is at fault. Scaling near 1080p to 1080p has quite high demands. As Anima123 already mentioned, in that situation it doesn't really make much sense to force image doubling. Basically you're doubling near 1080p to near 4K and then scale it back down to 1080p.

Originally Posted by leeperry

Is there any way to run xsbr and SR on my own via a PS script in my media player please?

Not with madVR, no. madVR does currently not support custom shaders which change video resolutions. This is planned for a future version, but it will take some time to get there.

Originally Posted by leeperry

I realize that you are the game creator and that you make the rules but post-processing settings are highly subjective, many apps come with a "simple/advanced settings" checkbox that has the former enabled by default. Could you for instance imagine Ahead Nero with 5 suboptions? Both newbies and experts want to be able to use it but they have different needs, one part of them likes as many options as possible and the other not. If at some point the latter wanna learn more about what's happening in the back, they can always go expert, learn and finetune it all optimally. Sometimes you don't know you need something till you try it.

At this point, you wanna dumb down mVR as much as possible in order not to scare new users away and all testers have to constantly complain, whine and try to prove their point that their favorite setting should stay(if it's not gone already)

I consider myself an expert, too, but I still want to get rid of as many options as possible. Even most experts do not want to adjust settings to different values for each movie. I know that some madVR users would prefer me to keep all options, but sorry, not going to happen. I do try to remove only options which I consider not useful enough to be worth the effort. And I know that some users may disagree about some of those decisions. But I make my decisions after listening to your feedback, and of course I have my own opinion, too, and I try to fuse them to get to the best possible end result. Anyway, you'll have to live with whatever decisions I make, sorry.

Originally Posted by leeperry

TBH I couldn't care less whether other users would prefer having HQ on or off, it's a free world and we all have different rigs, different tastes and different eyes. Since when is a "tendency" supposed to decide what looks best to me?

What looks best to you isn't always fixed, though. One day you love a noisy picture and ask for ways to add even more grain. The next day you want to get rid of as much noise as possible. And when testing the dithering algos, you changed your mind a few times, too. Just saying this so that you can understand better that I can't make decisions based on the feedback of one single user. Which options look good or bad can be a matter of taste, sometimes, but sometimes it also simply depends on with which content you tested. Some options look better with some content but worse with other content. I need to find options which look better for the majority of content and the majority of users. And that's what I'm trying to do. Furthermore, HQ downscaling is actually the scientifically and technically more correct solution. Actually, the non-HQ-downscaling implementation was broken. It did not work as it was originally intended. Still, you liked it better, which is your right. But several other users had a different opinion, *I* had a different opinion, and then add to that the scientific background and the fact that the non-HQ option was actually doing stuff wrong. All this combined led me to the decision to use HQ downscaling.

Originally Posted by leeperry

If I understood correctly, SR constantly compares and tries to find the best upscaled frame so they really only make sense in motion IMHO.

No. What you're describing is one method/implementation of SuperRes. But that's totally different to what Shiandow is doing. Shiandow's SuperRes strictly looks at each frame separately.

Originally Posted by leeperry

If this sample doesn't prove my point to your eyes then nothing will, really: http://forum.doom9.org/showpost.php?...ostcount=30138

This is a custom reencoded Anime sample with ringing artifacts in the source. I've not compared this sample with HQ downscaling on/off. But in any case, do you really want to base your decisions on an Anime reencoding which has EE artifacts in the source already? Of course testing such sources is useful, too, but only as one of many tests.

Originally Posted by leeperry

Anyway, 1.0 sharpness was way too sharp to me in .13 and if SR is bound to become a silly "SR" checkbox with zero subsetting then it sounds like I would be stuck with .13 forever or would have to run SR on my own in my media player(which is totally fine as I like to set all settings to my own personal subjective taste, I'm crazee like that sometimes).

There are new options to play with in build v0.88.16. Let me know what you think. I do value your feedback, but your feedback doesn't "overwrite" the feedback of other users.

Originally Posted by leeperry

Very nice, now you're calling me a nutcase agreeing with himself

My post was right under this one and if you look closely I did put a "^" character at the beginning in order to avoid a useless quote and hopefully making clear that I was referring to it:

It does sound like he'd try 87 should he be able to, the same way I'd love to try 37.

Neither he nor anyone else in the proximity of your original comment made a wish at all about a super-xbr sharpness knob. So there was nobody you could agree with there, except yourself...

Originally Posted by SithUK

I mostly watch 1080p content at 1080p resolution. Madvr operates using near neighbour by default I think when there is no scaling required. What other options could I try to improve performance with a 3d lut? There are no performance issues when I run without a 3d lut, but the main attraction to me is the calibrated rec709 gamut.

Try setting chroma upscaling to Bilinear.