madVR - high quality video renderer (GPU assisted)

[chros]

UHD->FHD chroma conversion bug introduced in 0.90.16 (0.90.15 works fine), from OSD:
- scale 0,0,3840,2160 -> 0,0,1920,1080
- chroma -> Catull-Rom
Where it shouldn't apply any chroma conversion in this case.

[70MM]

Im using version 0.89.17 and have some questions....
Im playing at 1080p > 1080p and using the JVC RS600 projector, video card 970

1. Is there a page I can look at that has an explanation for every setting please?

2. I play a lot of clips from Youtube which could be 720, 480 sometimes worse.... What are the best settings for getting these to look as good as I can please? Tidy up macroblocking, softness etc etc.... Is this area just in upscaling?

Thanks in advance...

[Clammerz]

Originally Posted by 70MM

1. Is there a page I can look at that has an explanation for every setting please?

As has been mentioned multiple times documentation most likely won't come until or near 1.0.
The best you'll get until then is probably this:
http://forum.doom9.org/showthread.php?t=171787

Originally Posted by 70MM

Tidy up macroblocking

If you looked you've probably noticed madVR doesn't have a deblocker (yet?) so you'll either need to preprocess with ffdshow, or make do.

Originally Posted by 70MM

softness etc etc.... Is this area just in upscaling?

Not sure what this really means, but you must have seen the "image enhancements" section by now.

[Warner306]

Originally Posted by 70MM

Im using version 0.89.17 and have some questions....
Im playing at 1080p > 1080p and using the JVC RS600 projector, video card 970

1. Is there a page I can look at that has an explanation for every setting please?

2. I play a lot of clips from Youtube which could be 720, 480 sometimes worse.... What are the best settings for getting these to look as good as I can please? Tidy up macroblocking, softness etc etc.... Is this area just in upscaling?

Thanks in advance...

Try the link in my signature. It is updated all the time with my own research. Most settings are covered.

[leeperry]

problem occurs only in windowed mode, all is well in FSE and that's with NNEDI3 luma doubling only as enabling it for chroma as well fixes the problem:

sample video and settings.bin available at https://mega.nz/#!D0IinBxJ!nv5b36Oqk...6YHjeBRh2bMCKc

I also got 720p videos that give the exact same problem but only in FSE this time.

[ryrynz]

Have had a number of issues changing queues in madVR over the years, this one kicked me out of the desktop and back into the sign in screen under Windows 10 after some graphics glitches
I changed the full screen and windowed mode frames presented in advance from 6 to 16 and the CPu and GPU buffers from 8/12 to 20/16 and then attempted to continue playing the video.

I had to reboot because the graphics driver wasn't stable after logging back in.

Seems the queues are simply too much for the old HD3000 to handle, those settings force Windows to do as I described every time I try to play something.
Setting them to 8/8/16/12 works fine.

[mzso]

I was testing profiles. I made it very simple. The first with sharpening the second without.
But I ran into issues with the hotkeys. First madvr didn't want to accept my shortcut falsely claiming it's invalid, because it used a character with a diacritical mark.
On second try I added alt+q and alt+w. Which were accepted but didn't work at all.

[madshi]

Originally Posted by clsid

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)

Makes sense.

Originally Posted by Neo-XP

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

crispen edges is not identical to FineSharp. Some of those options make no sense for crispen edges. Some may, but I don't really see the sense of exposing all that. Which of those do you consider important and why?

Originally Posted by 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.

Works for me. Are you sure that supersampling is really active? Can you make a screenshot of the OSD?

Originally Posted by 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.

On a quick check NNEDI3 still has some lines a bit more focused/sharp. But super-xbr may have slightly less aliasing in some situations.

Originally Posted by 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.

Sounds like a driver issue, but it's hard to say. Have you tried different GPU drivers for both Intel and NVidia GPUs? You could also try different media players, just to be safe. In any case, I think this is unlikely to madVR's fault, although I can't really say for sure.

Originally Posted by iSeries

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.

Seems to work fine on my PC, with supersampling and "automatically detect black bars" enabled, using a cinemascope HD video for testing.

Originally Posted by leeperry

I think AB is too strong in the new sxbr options

Meaning?

Originally Posted by 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.

Quality per performance isn't bad, although personally I prefer Lanczos over Spline in that performance area. Matter of taste, I guess.

Originally Posted by 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/...3/26/TrvLI.png
super xbr 100 anti bloat: https://www.ezphotoshare.com/images/...3/26/TrXIA.png
super xbr 100: https://www.ezphotoshare.com/images/...3/26/Trl1D.png

HD -> QHD
source: https://www.ezphotoshare.com/images/...3/26/TrZLg.png
spline3 AR: https://www.ezphotoshare.com/images/...3/26/TrJlx.png
super xbr 100 anti bloat: https://www.ezphotoshare.com/images/...3/26/TrSpB.png
super xbr 100: https://www.ezphotoshare.com/images/...3/26/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?

I think there's a small misunderstand: The option called "super-xbr, anti-bloating: 100" is *not* based on super-xbr 100. The anti-bloating filter was mainly written for sharpening, so it took some inventive tricks to make it work for super-xbr. Think of "super-xbr, anti-bloating xxx" as some special super-xbr configuration with different kinds of anti-bloating strength. Which means "super-xbr, anti-bloating: 100" has the highest anti-bloating strength, which makes it the softest of the anti-bloating super-xbr algorithms. If you find this one too soft, try the other anti-bloating variants. "super-xbr, anti-bloating: 25" should be the sharpest of the anti-bloating super-xbr algorithms.

Originally Posted by chros

UHD->FHD chroma conversion bug introduced in 0.90.16 (0.90.15 works fine), from OSD:
- scale 0,0,3840,2160 -> 0,0,1920,1080
- chroma -> Catull-Rom
Where it shouldn't apply any chroma conversion in this case.

Doesn't seem to happen on my PC. Screenshot of the OSD, please?

Originally Posted by leeperry

problem occurs only in windowed mode, all is well in FSE and that's with NNEDI3 luma doubling only as enabling it for chroma as well fixes the problem

Is this a new problem with v0.90.16? Or does this also occur with v0.90.15?

Originally Posted by mzso

I was testing profiles. I made it very simple. The first with sharpening the second without.
But I ran into issues with the hotkeys. First madvr didn't want to accept my shortcut falsely claiming it's invalid, because it used a character with a diacritical mark.
On second try I added alt+q and alt+w. Which were accepted but didn't work at all.

Maybe these are interpreted as menu commands? I tried with "Ctrl+Alt+1" as a test and that worked just fine for me.

[mogli]

Originally Posted by madshi

[...]
The option called "super-xbr, anti-bloating: 100" is *not* based on super-xbr 100.
[...]

On what is it based then?
Can they be compared like 'super-xbr anti-bloating 100' meaning 'super-xbr 150 minus the softening anti-bloating 100 does' or something?

[chros]

Originally Posted by madshi

Doesn't seem to happen on my PC. Screenshot of the OSD, please?

Strange. Thanks.

[Px]

Originally Posted by madshi

Thanks. I hope to have this fixed in the next build. If not, please let me know.

Looks like fixed, thanks

[iSeries]

Originally Posted by madshi

Works for me. Are you sure that supersampling is really active? Can you make a screenshot of the OSD?.

Sure. Using a 1.85.1 movie as an example (Ant Man), I have 'always - 2x supersampling' set in image doubling. First screenshot has 'automatically detect hardcoded black bars' selected, along with 'zoom small black bars away'. Doubling is being performed but no upscaling refinements have any effect (notice the render times). If I untick 'automatically detect hardcoded black bars' render times more than double and upscaling enhancements are clearly working (second screenshot).

http://imgur.com/Hli7UJI

http://imgur.com/MfuNVg5

On a 2.35.1 movie, with 'automatically detect hardcoded black bars' selected, along with zoom big black bars away completely so the picture fills the screen, upscaling refinements work fine along with doubling.

[Ver Greeneyes]

Originally Posted by madshi

Which anti-bloating strength do you prefer? And other than that AdaptiveSharpen sharpens more than Crispen Edges, do you see a quality difference between the two?

Sorry, I haven't played around with it to that extent yet. The video I was watching didn't seem very good for testing since the difference was pretty subtle. Today I played around with another video (640x360) and found that Crispen Edges allowed AdaptiveSharpen+AB to be a lot more effective - neither filter was as good on its own. But that's probably going to be overkill for other videos. I've just stuck with strengths of 1.00 and 100% for AB so far.

As for the difference between Crispen Edges and AdaptiveSharpen+AB, I think Crispen Edges has the drawback of emphasizing compression artifacts, whereas AdaptiveSharpen makes things look like a watercolor painting at higher strengths (AB does a lot to reduce this effect). I'm pretty sensitive to compression artifacts, at least on still images, so I tend to prefer AS at fairly low strength, but I'm not sure I could find a combination of settings that works well for all (or even a large subset of) the videos I watch. It's fun to play around with for low quality videos though!

[nevcairiel]

Originally Posted by chros

Strange. Thanks.

When you use Linear Light downscaling, this usually happens.
There is a new "trade quality of performance" option, called "scale chroma separately, if it saves performance", maybe try that?

[Georgel]

Originally Posted by huhn

and why don't you post a screen than?

Sorry, I don't seem to be able to post images.

Is it ok if I post link to dropbox files?

[Warner306]

I have a couple of examples of where anti-bloating removes detail that would be considered valid sharpening.

SD -> FHD
super-xbr100 + SuperRes (3)

Edit: Updated links to Google Drive.

In the first example, search for the headline "Rick Nielsen: no cheap collector" in the middle of the image. SuperRes does a good job of making the original letters more clear and easy to read. While anti-bloating returns these letters to a blurred smear.

The rest of the lettering, especially on the lower part of the guitar shows the same improvement and then degradation when anti-bloating is enabled.

Guitar
Guitar SuperRes (3)
Guitar SuperRes (3) + Anti-Bloating 100%

In the second example, look at the leather patches on Kata Mara's arm and the black guy.

Also, the objects in the shadows (top-right) become obscured when anti-bloating is enabled. SuperRes again slightly enhances the low-res original and anti-bloating effectively cancels this out.

Fantastic Four
Fantastic Four SuperRes (3)
Fantastic Four SuperRes (3) + Anti-Bloating 100%

So far, only my original test where SuperRes (3) was combined with crispen edges (1.0) did anti-bloating prove to be an effective countermeasure for SuperRes. I am thankful to see that SuperRes, at the right strength, does not oversharpen the image. This has completely changed my settings preferences.

[chros]

Originally Posted by nevcairiel

When you use Linear Light downscaling, this usually happens.
There is a new "trade quality of performance" option, called "scale chroma separately, if it saves performance", maybe try that?

Hmm, thanks.
- 'scale in linear light' under image downscaling is unchecked in the profile (checking/unchecking doesn't make a difference)
- if I check 'scale chrome separately, if ...' then indeed chroma conversion vanishes (so maybe this new option brought this bug, if it's a bug)

My Zoom control settings for 4k (maybe it also affects something):
- 'disable scaling if image size changes by only' checked: 50 lines or less
- 'move subtitles' checked: to bottom of the screen/window
- 'auto detect hard coded black bars' UNchecked

[har3inger]

Superres with antibloat at or above 100% completely eliminates the aliasing/blobbiness/chunkiness SR introduces to low resolution content. Unfortunately, it also makes it appear that superres is doing almost nothing except introduce artifacts on hardcoded subs and similar scene details.

For low-res bad quality content, SXBR AB25 looks quite similiar to nnedi64. There's a bit less sharpness, but similar reduction in source artifacts. Considering the cost, that's pretty damn good. Any higher AB and SXBR starts looking too soft. Using AB on SXBR followed by SR with AB checked can tend to produce odd looking images.

I'm not able to take or upload screenshots for a while unfortunately, so I can only leave these impressions here. It'd be really great if someone could try to reproduce and see if they notice similar issues.

What quality tradeoffs are present when we check "scale chroma separately"?

[nevcairiel]

Originally Posted by chros

- if I check 'scale chrome separately, if ...' then indeed chroma conversion vanishes (so maybe this new option brought this bug, if it's a bug)

Well sounds to me like this option is designed to allow just what you want, scale chroma straight to the target (which in this case is no scaling at all). But madshi can say if it works as intended.

[Warner306]

Originally Posted by har3inger

Superres with antibloat at or above 100% completely eliminates the aliasing/blobbiness/chunkiness SR introduces to low resolution content. Unfortunately, it also makes it appear that superres is doing almost nothing except introduce artifacts on hardcoded subs and similar scene details.

For low-res bad quality content, SXBR AB25 looks quite similiar to nnedi64. There's a bit less sharpness, but similar reduction in source artifacts. Considering the cost, that's pretty damn good. Any higher AB and SXBR starts looking too soft. Using AB on SXBR followed by SR with AB checked can tend to produce odd looking images.

I'm not able to take or upload screenshots for a while unfortunately, so I can only leave these impressions here. It'd be really great if someone could try to reproduce and see if they notice similar issues.

What quality tradeoffs are present when we check "scale chroma separately"?

Sure, I can agree about the reduction in aliasing. But it is doing this by blurring the aliasing a bit. This also impacts other parts of the image, though, and the overall effect can be detrimental when the image is softened when it is not oversharpened.