madVR - high quality video renderer (GPU assisted)

[SweetLow]

Originally Posted by madshi

Seems to work fine here. Are you sure?

Yes.
With 3 devices in list (in that order):

Samsung LE40B530 HDMI
LG Display LP156WF4-SPB1
Dell 2707WFP

when attached 1 and 2 - variable display="Samsung LE40B530 HDMI" always
when attached 2 and 3 - variable display="LG Display LP156WF4-SPB1" always

this script (in group created for dithering):
if (display = "Samsung LE40B530 HDMI") "Samsung"
elseif (display = "LG Display LP156WF4-SPB1") "LG"
elseif (display = "Dell 2707WFP") "Dell"
else "None"

when only one device is active - variable is set right ("None" profile is never set actual)

madVR right to detect actual device (devices use different bitdepth in Exclusive FS so it's easy to check in debug OSD, correctly detect exact display rate, etc), but not set right variable.

[madshi]

Originally Posted by leeperry

My real world tests also made me prefer St4/Sh2

Originally Posted by aufkrawall

Sharpness of 2 is definitely most useful.

Good to hear we're all on the same page. So I've removed the sharpness option now and set it to the algo which produced the best results in my test (near to sharpness value 2, but not identical).

Originally Posted by aufkrawall

With a sharpness of 100, you get very fat lines compared to NNEDI3

Maybe the new line thinning algorithm in v0.89.14 helps?

Originally Posted by aufkrawall

This source however is already a bit aliased without any scaling (it has some ringing too).
I don't know about such fines structures. Maybe it is correct that they look aliased when they are upscaled, since this could be how the image would look like if it was created in that size in the beginning?
Maybe it's just a problem with very tiny pictures (or huge scaling factors).

I'm not sure. Of course we could just stick to a 0.66 radius and wait for users to complain about aliasing problems with real world material.

Originally Posted by aufkrawall

However, with content "in the wild", I've found the thickening of LL SR more annoying than the brightness increase of gamma light in the meantime.

Yes, same here.

Originally Posted by aufkrawall

Yes. For this reason, I now use SR with the default values instead of strength 2 and sharpness 3. They end up mostly with the same sharpness and thanks to your optimizations, 3 passes are affordable now.

[...]

Hm, but there could be the case when you want to use SuperRes, but can only afford 2 passes, and still you don't want to miss out on some additional sharpness. In this case, a sharpness of 3 could still make sense.

Strength is not the same as passes. Strengths 2 and 3 have the same performance in madVR.

Originally Posted by onyx76

Hello, anyone have any suggestions to improve my settings?

• Double Luma: 1.2x or greater - NNEDI3 64 Neurons
• Double Chroma: 1.2x or greater - NNEDI3 32 Neurons
• Quadruple Luma: 2.4x or greater - NNEDI3 64 Neurons
• Quadruple Chroma: 2.4x or greater - NNEDI3 32 Neurons

Chroma doubling/quadrupling is almost useless. I'd disable it, to save performance for other processing algorithms.

Originally Posted by onyx76

What is better NNEDI3 + Adaptive Sharpen or NNEDI3 + SuperRes?

I prefer SuperRes, but you decide.

Originally Posted by Ver Greeneyes

That would work for DVDs, but I watch a lot of 1080p Youtube videos (1080p and 720p are both very common for Youtube and twitch). Anyway, I think it would be of more use to you if I just got you samples where it unexpectedly fails - I'll enable it and keep an eye out

Oh, ok. Well, for weird PC games with lots of black in them, there's a good chance my black bar detection might not work reliably. If I try to optimize for that, I might end up dramatically slowing down detection speed for normal movies, in order to avoid false positives with PC game recordings. Is there maybe some other way you can identify those Youtube/twitch videos via profiles?

Originally Posted by nlnl

The problem: When source is video (29,97) and "force video mode' option is on, display does not switch to 59 rate, but swtches to 23.
If I set to 'video mode' using keyboard shortcut display switches.

Hmmmm... Which decoder? Software decoding? DXVA copyback? Native DXVA? Does this problem occur with all videos or just some?

Originally Posted by markanini

It was Intel QS decoding in LAV filters. Switching to DXVA2 or disabling hardware decoding solves the problem.

Oh, good to hear it was (seemingly) not my fault.

Originally Posted by agustin9

madshi, quick question, could Error Diffusion be doubling the cpu usage? Thanks!

I don't know for sure. If madVR is spending that CPU usage then it's done in a way that (although it pushes CPU consumption higher) doesn't stop other tasks from being executed. So it's probably not a big problem.

Originally Posted by SweetLow

when only one device is active - variable is set right

Ah, I understand. I only ever tested with 1 active device. Will add this to my to do list, maybe for next week end, not sure.

[madshi]

madVR v0.89.14 released

http://madshi.net/madVR.zip

* added new "sharpen edges" algorithm
* added new "crispen edges" algorithm (tamed FineSharp)
* added new "thin edges" algorithm
* added new "enhance detail" algorithm
* removed SuperRes "radius" option (set to 0.66)
* removed SuperRes "sharpness" option (set near to 2)
* removed LumaSharpen "radius" option (set to 1.0)
* removed LumaSharpen "clamp" option (set to 0.035)
* removed "apply SuperRes first" option (set to true)
* updated AdaptiveSharpen to 2015-11-05 (experimental) version
* optimized "video size has changed" notifications to media player
* optimized aspect ratio overriding API behaviour

You know, I was planning to slow down development for a while, but I just couldn't hold back, so here you go. Maybe I'll manage to concentrate on commercial work now, we'll see...

Try the 4 new algos and let me know what you think, and how they compare in your opinion to SuperRes/AdaptiveSharpen/LumaSharpen. FineSharp lovers, don't cry, the new "crispen edges" algo is a tweaked FineSharp version which doesn't ring and isn't as sensitive to noise, anymore. FineSharp haters, give the "crispen edges" algo a chance, maybe you'll like it now.

Do we still need AdaptiveSharpen and/or LumaSharpen?

[aufkrawall]

Originally Posted by madshi

Maybe the new line thinning algorithm in v0.89.14 helps?

Sounds like a wonderful thing.
Edit: Oh, there it is already.

Originally Posted by madshi

I'm not sure. Of course we could just stick to a 0.66 radius and wait for users to complain about aliasing problems with real world material.

I can see some tiny bits of aliasing with SR & leeperry's lighttower image when quadrupling it from 540p to WQHD. However, this seems to be again because the structure is so thin. With LL SR, the dark railing is bigger and shows as good as no aliasing.
But even without SuperRes and just NNEDI3 64, some aliasing can be "guessed", it just seems blurrier.

Originally Posted by madshi

Strength is not the same as passes. Strengths 2 and 3 have the same performance in madVR.

Oops, then I maybe made one mistake in the past when testing around different settings. Weird, but shit happens.

Originally Posted by madshi

I prefer SuperRes, but you decide.

I agree, SuperRes takes away that "oil painting" look of NNEDI3 to a more naturalistic one (like super-xbr, but without the drawbacks).

[Budtz]

Thx for the new version.

I have used finesharp up til now. Can I use the strength-setting from finesharp with the "crispen edges" algorithm (tamed FineSharp)? Meaning setting it to 1 og 1,5 or 2 i will get roughly the same amount of sharpening as I got from finesharp with the same setting.

[Budtz]

BTW is there any consensus now on the best sharpening settings? I have a fast system and pretty much only watch blu-ray quality material.

[aufkrawall]

Fantastic new features.
"sharpen edges" even doesn't increase brightness with a strength of 4 and "thin edges" is a great addition for SuperRes LL or super-xbr.
Very impressive, madshi. Kudos!

[JarrettH]

Originally Posted by madshi

madVR v0.89.14 released

* added new "sharpen edges" algorithm
* added new "crispen edges" algorithm (tamed FineSharp)
* added new "thin edges" algorithm
* added new "enhance detail" algorithm

Let the madness and testing begin

[truexfan81]

loving the "enhance detail" does wonders for those old xvid files

[foozoor]

@madshi

- Is there madvr presets like on mpdn? Too much options for noobs.
- Could we get hlsl shader version of your tamed finesharp?
- I think you can remove lumasharpen and adaptive-sharpen too.

Thanks.

[madshi]

Originally Posted by aufkrawall

I can see some tiny bits of aliasing with SR & leeperry's lighttower image when quadrupling it from 540p to WQHD. However, this seems to be again because the structure is so thin. With LL SR, the dark railing is bigger and shows as good as no aliasing.
But even without SuperRes and just NNEDI3 64, some aliasing can be "guessed", it just seems blurrier.

So - should we stick with 0.66 for now?

Originally Posted by Budtz

I have used finesharp up til now. Can I use the strength-setting from finesharp with the "crispen edges" algorithm (tamed FineSharp)? Meaning setting it to 1 og 1,5 or 2 i will get roughly the same amount of sharpening as I got from finesharp with the same setting.

I've changed the settings range so that 1.0 is default everywhere (for the 4 new algos). FineSharp had a default of 2.0. So basically with the same "strength" the new algo is twice as strong. However, due to the anti-ring filter and noise modifications, the new algo loses some strength, so you have to try for yourself which new strength you need/like.

Originally Posted by Budtz

BTW is there any consensus now on the best sharpening settings? I have a fast system and pretty much only watch blu-ray quality material.

I've *just* released 4 new sharpening algos, and you're already asking for a consensus? How would that be possible?

Originally Posted by aufkrawall

Fantastic new features.
"sharpen edges" even doesn't increase brightness with a strength of 4 and "thin edges" is a great addition for SuperRes LL or super-xbr.
Very impressive, madshi. Kudos!

Glad to hear your first impression is positive. Hope to hear more detailed feedback when you've had some time to do some in depth testing. E.g. since you're an AdaptiveSharpen user, do you prefer AdaptiveSharpen over "sharpen edges"? Or the other way round? And do you like super-xbr better now, in combination with "thin edges"?

Originally Posted by truexfan81

loving the "enhance detail" does wonders for those old xvid files

Glad to hear that - but isn't there too much noise/artifacts in those xvid files? The "enhance detail" algo has the side effect of enhancing noise/artifacts together with image detail, so I thought it would mostly be suitable for clean HD sources, only.

-------

Forgot to mention: The new algos are not fully performance optimized yet. So don't put too much weight on their performance. Performance will probably improve at some point in the future.

[truexfan81]

Originally Posted by madshi

Glad to hear that - but isn't there too much noise/artifacts in those xvid files? The "enhance detail" algo has the side effect of enhancing noise/artifacts together with image detail, so I thought it would mostly be suitable for clean HD sources, only.

it does a bit, but i can ignore it

[Warner306]

Is is possible to add some documentation on what the sharpening shaders are supposed to do? The titles are less than useful.

Right now, the performance seems as follows:

Low: Enhance Detail, LumaSharpen

Medium: Crispen Edges, AdaptiveSharpen

High: SuperRes, Sharpen Edges, Thin Edges

[obieobieobie]

I get a black screen in madvr after I updated to 0.89.14 from 0.89.13. I did not change any parameters, I just unpacked the new zip over the old install.

[madshi]

Originally Posted by foozoor

Is there madvr presets like on mpdn? Too much options for noobs.

The default options should be ok for a start. Simply leave all options that you don't understand at their default values, and just play with those options that make sense to you.

There are no presets atm. Presets, an expert/noob switch, documentation/manual, and all that kind of stuff will come when I release v1.0, but not before.

Originally Posted by foozoor

Could we get hlsl shader version of your tamed finesharp?

No, sorry.

Originally Posted by truexfan81

it does a bit, but i can ignore it

Ah, ok.

Originally Posted by Warner306

Is is possible to add some documentation on what the sharpening shaders are supposed to do?

I had hoped the names would be self explaining. "sharpen edges" is just another line/edge sharpener, somewhat similar to "LumaSharpen" and "AdaptiveSharpen". However, I like "sharpen edges" better because it doesn't have the typical fat/bloated look of other line sharpeners. "crispen edges" tries to make high-frequency edges "crispier", meaning the image doesn't just get sharper, it gets a more high-res look. "thin edges" tries to make edges, lines and even full image features thinner/smaller, which is especially useful after upscaling, because upscaling often bloats lines and features up a bit. All the above algos concentrate on edges and leave flat image areas mostly untouched. The "enhance detail" algo is pretty much the opposite. It concentrates on making faint image detail (and noise and artifacts) in flat image areas more visible, while mostly ignoring lines/edges.

Originally Posted by obieobieobie

I get a black screen in madvr after I updated to 0.89.14 from 0.89.13.

Does it happen with all videos or just with some specific? Try videos with different properties, e.g. MPEG2 vs h264, interlaced vs progressive etc. Maybe I can see what's going on if you create a debug log.

[QBhd]

@madshi:

Is there any drawbacks to enabling all 4 of the new algo's? I have not tested yet, but they all sound great so I just clicked all 4. I will probably be watching some stuff tonight, so I will try and report findings after that. Just curious if madVR will imploded after enabling all 4 LOL

QB

[aufkrawall]

Originally Posted by madshi

So - should we stick with 0.66 for now?

Well, with your new sharpen algorithm things have changed a bit.
Could you please bring back the radius option for a build or two? I suppose with higher radius and new "sharpen edges", very decent results could be achieved.
With "thin edges" the aliasing gets more visible, so there's really a good reason for higher radius when combining it with SR LL.

Originally Posted by madshi

Glad to hear your first impression is positive. Hope to hear more detailed feedback when you've had some time to do some in depth testing. E.g. since you're an AdaptiveSharpen user, do you prefer AdaptiveSharpen over "sharpen edges"? Or the other way round? And do you like super-xbr better now, in combination with "thin edges"?

I tested three images yet:
-leeperry's lighttower
-your FineSharp testimage
-my infamous cartoon picture
I must say that I'm very convinced of the results so far.
With the lighttower picture (1080p -> WQHD postresize) it's very noticeable that the brightness doesn't increase, unlike with Adaptive Sharpen. It also seems to look rather naturalistic, Adaptive Sharpen with high strength has that "oil painting" look. However, "sharpen edges" can't be set that strong. But such high values aren't needed anyway.
The effect in cartoons is not very strong, but here does line thinning the main job anyway. "sharpen edges" shares the positive trait of Adaptive Sharpen to not boost ringing, which makes it also a good choice for non-optimal sources.

I haven't tested the new features is image enhancement yet, but to me it seems they could totally replace Lumasharpen and Adaptive Sharpen.

[ryrynz]

Originally Posted by madshi

* added new "thin edges" algorithm

Whaaaaaaaa

Is this an where near as good as awarpsharp?

Man I need to do some serious comparisons..

[madshi]

Originally Posted by QBhd

Is there any drawbacks to enabling all 4 of the new algo's?

No drawbacks at all. They don't seem to negatively impact each other.

Originally Posted by aufkrawall

Well, with your new sharpen algorithm things have changed a bit.
Could you please bring back the radius option for a build or two? I suppose with higher radius and new "sharpen edges", very decent results could be achieved.
With "thin edges" the aliasing gets more visible, so there's really a good reason for higher radius when combining it with SR LL.

Oh well... I'm going to bed now, maybe tomorrow.

Originally Posted by aufkrawall

I tested three images yet:
-leeperry's lighttower
-your FineSharp testimage
-my infamous cartoon picture
I must say that I'm very convinced of the results so far.
With the lighttower picture (1080p -> WQHD postresize) it's very noticeable that the brightness doesn't increase, unlike with Adaptive Sharpen. It also seems to look rather naturalistic, Adaptive Sharpen with high strength has that "oil painting" look. However, "sharpen edges" can't be set that strong. But such high values aren't needed anyway.
The effect in cartoons is not very strong, but here does line thinning the main job anyway. "sharpen edges" shares the positive trait of Adaptive Sharpen to not boost ringing, which makes it also a good choice for non-optimal sources.

I haven't tested the new features is image enhancement yet, but to me it seems they could totally replace Lumasharpen and Adaptive Sharpen.

Thanks for the feedback. Sounds promising!

Originally Posted by ryrynz

Whaaaaaaaa

Is this an where near as good as awarpsharp?

Man I need to do some serious comparisons..

Please do and let us know your results!

[obieobieobie]

Originally Posted by madshi

Does it happen with all videos or just with some specific? Try videos with different properties, e.g. MPEG2 vs h264, interlaced vs progressive etc. Maybe I can see what's going on if you create a debug log.

I played two files:

1.mkv shows the video just fine in 0.89.14
2.mkv black screen in 0.89.14

here's the log:

http://s000.tinyupload.com/index.php...13829934374187