madVR - high quality video renderer (GPU assisted)

[nijiko]

Originally Posted by michkrol

There is no way to check it with profile rules, all the variables you can check are listed here: http://forum.doom9.org/showthread.php?p=1271417

If you want to see what scaling (and doubling) algorithms are used, check debug OSD (Ctrl+J).
If you want to see which profiles are activated, check settings - the active profile's name is bold. Take note you must sometimes close and re-open the settings dialog to see active profiles, since it's not updated in real-time.

I know, I use madVR for many years.
These days, I found when using Super-xBR, it's no need to use Jinc (drop frames), but if Super-xBR is not actived, Jinc is demanded.
So, I want to find a way to judge these cases.

[madshi]

For those of you still having some weekend hours left for playing, here's a pre-release version of the next madVR build:

http://madshi.net/madVR8814pre.zip

It has the new SuperRes version inside, plus a new chroma upscaler, and the AdaptiveSharpen algo. So some new things to play with. Official version will come soon.

[fairchild]

Originally Posted by madshi

For those of you still having some weekend hours left for playing, here's a pre-release version of the next madVR build:

http://madshi.net/madVR8814pre.zip

It has the new SuperRes version inside, plus a new chroma upscaler, and the AdaptiveSharpen algo. So some new things to play with. Official version will come soon.

I'm trying out the AdaptiveSharpen as image enhancement only on HD content and at the default 0.5 it has a very good looking sharpen that is noticeable. I'll leave it up to the experts to check it with blown-up test images to see if there are any artifacts.

[James Freeman]

AdaptiveSharpen definitely does a good job for first impression.
I noticed that it does not touch the already sharp detail thus no halo or ringing is visible.
With only one knob ! Cool.

IMO only:
The new Bilateral chroma scaler is so bad it belong between Nearest Neighbor and Bilinear.
To me it looks like a slightly (even that is too much) better (or worse?) Bilinear.
Zoom in to a red object and switch between chroma upscalers.

[madshi]

Replies to all other emails coming soon.

Originally Posted by James Freeman

The new Bilateral chroma scaler is so bad it belong between Nearest Neighbor and Bilinear.
To me it looks like a slightly (even that is too much) better (or worse?) Bilinear.
Zoom in to a red object and switch between chroma upscalers.

Screenshot/sample, please? Bilateral is very different in concept. In some cases it's much better than anything else. In other situations it might be worse.

[6233638]

Originally Posted by Warner306

I see 4:4:4 in purple lettering with no blinking. Is this an assumed pass?

There is an updated test pattern with examples here:

http://forum.doom9.org/showpost.php?...ostcount=25082

@Madshi perhaps you could update the image on your server?

[James Freeman]

Originally Posted by madshi

Screenshot/sample, please? Bilateral is very different in concept. In some cases it's much better than anything else. In other situations it might be worse.

http://www.mediafire.com/download/w6..._Comparison.7z

EDIT: Which cases madshi?

[aufkrawall]

Biliteral chroma scaling looks totally terrifying with my cartoon example.
Chroma bleeding, aliasing etc.

Edit: Hm, retrospectively I'd say even super-xbr chroma looks terrible compared to Jinc due to problems with ringing.
Edit 2: Forget that, anti-ringing filter wasn't activated.

[Shiandow]

Originally Posted by James Freeman

The new Bilateral chroma scaler is so bad it belong between Nearest Neighbor and Bilinear.
To me it looks like a slightly (even that is too much) better (or worse?) Bilinear.
Zoom in to a red object and switch between chroma upscalers.

Originally Posted by aufkrawall

Biliteral chroma scaling looks totally terrifying with my cartoon example.
Chroma bleeding, aliasing etc.

Oh dear, seems that the bilateral filter was even more experimental than I thought. If you want to help with the development you could try doing the following:

Go to the "legal-stuff\Shiandow" folder and extract the "Bilateral.hlsl" file from the "RenderScripts.zip". Then try fiddling with the following lines:

#define acuity 15.0
#define radius 0.66
#define power 3.0

Things you could try are: lowering acuity to 9.0 (or even more), or setting power to 0.5 and acuity to 100.

[aufkrawall]

Will give it a try.

However, everyone should be easily able to see it, 52kb sample download:
http://www39.zippyshare.com/v/cDstAbaj/file.html

I scale it up to WQHD with NNEDI3 64 for luma.

I suppose SuperRes for chroma is updated as well?
It doesn't seem to vanish contoures anymore, instead they look more clearly defined now.

[aufkrawall]

AdaptiveSharpen has a very strong effect, even with strength 0.1 the effect is very visible as an image inhancement (especially with that upscaled cartoon sample).
Hm, maybe it's a bit less destructive than FineSharp. Will need to take a closer look with blocking artifacts.

Now that SuperRes for chroma doesn't seem to vanish contoures anymore, it seems it can be enabled with default values without any disadvantage. Chroma looks a little clearer with it with Jinc AR.
Btw: With that cartoon example, Jinc also looks clearer than super-xbr. Look at the flowers of the headband.

SuperRes with low strength (1 pass, strength 0.3, softness 0.15) "magicially" seems to be able to repair structures. They look less aliased and can be more easily identified.

[Anima123]

Originally Posted by aufkrawall

SuperRes with low strength (1 pass, strength 0.3, softness 0.15) "magicially" seems to be able to repair structures. They look less aliased and can be more easily identified.

Are you talking about Chroma or Luma here with SuperRes?

[madshi]

Originally Posted by JarrettH

One other thing: did you notice those blue artifacts in #3?

Originally Posted by viewer

sample here:
https://mega.co.nz/#!bR8mjSCT!7ki_Lv...AMxXB1mYPdTxvs

Thank you. This is a good sample.

Those blue artifacts are caused by chroma ringing. They occur most strongly with super-xbr with AR disabled. Next one in the list if Lanczos4 without AR. Then later comes Lanczos3/4 with AR, because the AR doesn't remove all ringing.

However, already v0.88.13 removed the problem for super-xbr, because v0.88.13 has a very strict AR algorithm implemented for super-xbr chroma upscaling. So no blue artifacts in viewer's sample when using super-xbr for chroma upscaling with v0.88.13 (or newer), if AR is enabled.

Originally Posted by SecurityBunny

Madshi, I have another case of queues not filling all the way when toggling between fullscreen, windowed, and fullscreen again. It is the render queue that drops (0/1-2 from 7-8), which subsequently drops the present queue fill rate. Just noticed since updating. I have to pause playback for a second for queues to fill all the way up again.

Going into fullscreen initially, queues are fine. The issue occurs when going fullscreen, then windowed, then back to fullscreen again. Please let me know if you need me to debug with a special file to help narrow down where the problem may be.

D3D11 10-bit
MPC-HC 1.7.9.25 x64
MadVR 0.88.13
Nvidia GeForce 353.30

Originally Posted by ryrynz

Yes, display mode switcher is currently off and refresh is already 60Hz.

Changing prepresented frames for Exclusive mode has no effect unless I set it to 1. If I set it to 2+ I get glitches on the screen (part flashes of previous frames) even though no glitches or dropped frames show on the OSD.
If I change prepresented frames to 1, everything seems to work fine.

Can you guys please first do some tests and collect some more data? E.g. does it only occur in 10bit output mode? Or also in 8bit output mode? Does the screen refresh rate matter? Do you have any other software running besides the media player? Please make sure you close everything else, even 3rd party background services, just to be safe.

Originally Posted by VenomousNinja

Activating use of Direct3D 11 in MadVR on my setup makes my secondary monitor connected via HDMI (LG TV) to do weird things during avi payback. GTX970, Win 7 x64, latest driver 353.30, MPC-BE. Exclusive full screen. Monitor just goes black 2-3 times during video playback for a second or two.I'd say every 20-25 mins. Sometimes it's just had been stuck in some kind of signal loss loop - black-video-black-video. Even after closing video player (MPC-BE) it's still remains in on - off loop. I've deselected dx11 and usinf DX9 for now. No mystic happens. This kind of stuff started to happen in the last 2-3 build of MadVR. Possibly it's nvidia new drivers...

This is probably caused by some D3D11 internal stuff, or by the GPU drivers. Probably nothing I can do about it.

Originally Posted by XMonarchY

I tested FineSharp on several sources and each time I could see obvious ringing using FineSharp in image enhancements and upscaling refinement. LumaSharpen does not SEEM to do that at default settings, so it is beyond me how people find FineSharp superior to LumaSharpen...

Can I get some screenshots/samples, please? And please not some 200x100 ultra low quality sources.

Originally Posted by XMonarchY

I think its a good idea to update the OP with new screenshots of how madVR improves image quality vs. other renderers. Something with NNEDI3 + SuperRes + whatever else would be good. madVR 0.5 is very VERY old and its gotten 100x better since then!

Yeah, might make sense, but doing that takes time, and I prefer working on madVR instead.

Originally Posted by rack04

Not sure if it's a bug or my system not keeping up but I get dropped frames at scene transitions with the trade quality for performance setting "don't render frames when fade in/out is detected" unchecked.

Try increasing the GPU queue size, that usually helps.

Originally Posted by Eyldebrandt

Finesharp is just about destruction imo.

I agree with you only as far as "thinning" goes. The main FineSharp is quite good, IMHO. Might only work for good sources, though.

Originally Posted by Dogway

I understand that the ideal situation (for dithering, etc) should be to set up Nvidia settings as "Full RGB" (do nothing), and then madVR to "TV Levels" for the TV display, the problem is that with "Full RGB" my desktop looks wrong (clipped).

Then set everything to PC/Full RGB, including your display.

Originally Posted by James Freeman

After some reading, can I say that overlay mode solves all the issues with windows DWM because it is completely bypassed?

What issues with DWM? There are no issues in Windows 8.1, as far as I'm aware.

Originally Posted by James Freeman

What about DX11 10bit overlay then? Is it possible?

Not possible, AFAIK.

Originally Posted by iSunrise

Since you neither share your settings (Finesharp) vs. LumaSharpen defaults nor do you provide screenshots to compare, it's a bit complicated to make meaningful conclusions without some hard facts.

There are way too many soft facts in this thread, which will ultimately lead to a product that could be better, but isn't, because people don't provide screenshots which show "what they see".

madshi can only make "good choices" with good quality feedback from this thread.

You just have to press "print screen" on your keyboard when not in fullscreen exclusive (windowed mode) at the same frame with the madVR settings you can change and will show up in realtime in your mediaplayer, upload it to some imagehost and post them here together with a short explanation "what you see". Takes less than 10 minutes to do and helps madshi as well as everyone else that uses madVR.

Fully agreed !!!

Originally Posted by vivan

And so what?
Even when there's no processing, even when there's no scaling - there's still chroma that needs to be scaled and color conversion to be performed.

Exactly.

Originally Posted by James Freeman

A typical video (mkv, Blu ray, DVD, you name it) is stored in limited range 16-235.
Selecting Full range in madVR WILL stretch it to 0-255 and will require dithering to prevent banding.

You can download the AVS709HD calibration patterns (all in video 16-236 range) and play the Grey Ramp video test pattern.
Now Disable dithering to see banding if madVR is set to Full range.
BUT, when you select Limited range in madVR and Disable dithering you will not see any difference at all with this pattern because madVR is not stretching the original range.

It is true that with a prefect color neutral grayscale, letting madVR render to 16-235 means that madVR doesn't have to dither. However, even black & white movies are usually not 100% color neutral. They often have a slight color tint. And as soon as chroma is anything but exactly color neutral, madVR instantly has to dither. Which is the case for 99.99% of all movies out there. And once madVR dithers, it shouldn't matter too much whether you configure madVR to output PC or TV levels.

Yeah, sure, in theory having madVR output 16-235 should be slightly better. But in real life I don't think there's much of a difference between 16-235 and 0-255 output. Of course 16-235 allows the preservation of BTB/WTW, which might be a plus because some displays might be calibrated to show a tiny tiny bit of BTB/WTW or something...

Originally Posted by lanzorg

FineSharp is really too bad, madvr needs implementation of LimitedSharpenFasterMC and UnsharpHQ from the avisynth world.

I don't have the time to convert the AviSynth script to D3D pixel shaders. The problem already starts with that I don't even fully understand what the script does. "impand", "expand", "lutxy", I've no idea what those do exactly or how to implement them. It sure would be possible to research all that, but it would cost a lot of time. Maybe some day. Or maybe someone else will step in, as -Vit- did with FineSharp.

If you don't like FineSharp, I'd like to see some screenshots/samples which show that it's "bad". Good quality sources, please.

Originally Posted by omarank

I don’t think super-xbr can replace Jinc as the best medium speed chroma upscaler, because Jinc is more natural looking. I prefer suber-xbr chroma (sharpness 100) over Jinc AR chroma in two instances, when the source is rather soft or the image is being downscaled by a large factor using CR AR LL.

Samples/screenshots, please?

FWIW, v0.88.13+ is better quality than v0.88.12.

Originally Posted by omarank

For image doubling, super-xbr is good, but NNEDI3 is much superior.

Yes, that's true. But the difference shrinks if you use the new SuperRes with enough passes (if you like the resulting look).

Originally Posted by lanzorg

Is it possible to change the madVR preferences from the command line?
I create a little installer and I need to enable smoothmotion and directx11 based on user choice.

Not from the command line, but you can write a little tool which loads the madVR filter and then calls the interfaces documented in this header:

madVR installation folder\developers\interfaces\mvrInterfaces.h

Originally Posted by James Freeman

AdaptiveSharpen definitely does a good job for first impression.
I noticed that it does not touch the already sharp detail thus no halo or ringing is visible.
With only one knob ! Cool.

I like the "one knob only" very much. Personally, I prefer the look of FineSharp, though.

Originally Posted by 6233638

There is an updated test pattern with examples here:

http://forum.doom9.org/showpost.php?...ostcount=25082

@Madshi perhaps you could update the image on your server?

Done, thanks.

[madshi]

madVR v0.88.14 released

http://madshi.net/madVR.zip

* updated to latest Shiandow SuperRes version
* added Bilateral chroma upscaling algorithm
* added AdaptiveSharpen to "image enhancements" and "upscaling refinement"
* improved luma anti-ringing quality for super-xbr image doubling
* super-xbr chroma upscaling now supports enabling/disabling anti-ringing
* added super-xbr sharpness options of 25 and 150
* ca. 30% performance boost for super-xbr image doubling
* ca. 15% performance boost for super-xbr chroma upscaling
* ca. 30% performance boost for FineSharp
* removed "thinning" parameter from FineSharp, now always set to 0
* added various new madTPG APIs

I consider the FineSharp feedback phase to be (mostly) finished. The interface will be simplified a bit, away from the now cumbersome edit/up/down control. Maybe it will be slider or something, we'll see. I'll decide on the range and number of "steps" after reviewing all your feedback.

Now let's move on to the new SuperRes algorithm. I'm talking about the upscaling refinement algorithm for now, *not* the chroma upscaling SuperRes. Let's concentrate on upscaling refinement for now. So I would like to have your feedback about the following questions:

First the setup:

* Please setup your media player to do exactly 200% enlargement. E.g. in MPC-HC choose "Double Size". This is important, because it helps us all comparing under the same circumstances.

* Please disable all sharpening algorithms (FineSharp, LumaSharpen, AdaptiveSharpen).

* Please choose one of the following algorithms for image doubling/upscaling and in your feedback report please state which one you selected:
- Catmull-Rom AR
- NEDI
- super-xbr
- NNEDI3

Questions:

1) Which image upscaling/doubling algorithm do you like to use with SuperRes and why?
2) Which values do you like for "strength" and "softness". Please note that the default values are strength=1.0, and softness=0.0. And these *may* be the best values. But you can still try other values to see if you like them more.
3) The "use HQ downscaling" option changes the overall "look" of SuperRes a bit. Which look do you prefer? Please note that with the option turned on, you may have to increase the number of passes, because the SuperRes effect is slightly less intense with this option turned on.
4) How many passes should be used as default?

Please don't put too much weight on performance. There is room for performance improvements. For now I'm mostly interested in discussion about quality.

Thanks!!

[iSunrise]

Thanks for the new test build madshi.

I did a very short (to be continued) test between the new Adaptive Sharpen and Finesharp.

The sharping effect is about equal when you have Finesharp at 1.0 and Adaptive Sharpen at 0.3.

From what I can see, Adaptive Sharpen has a lot less ringing (which may or may not be a good thing, not sure yet), but it still does a very good job digging up blurred details very nicely. Very interesting sharpener to say the least.

I need to do some further tests though, not sure what looks more natural/accurate to my eyes this late in the day. Will do some tests and some comparisons this week and report back.

[SecurityBunny]

Originally Posted by madshi

Can you guys please first do some tests and collect some more data? E.g. does it only occur in 10bit output mode? Or also in 8bit output mode? Does the screen refresh rate matter? Do you have any other software running besides the media player? Please make sure you close everything else, even 3rd party background services, just to be safe.

It seems to only occur when using D3D11 with 10bit output. D3D11 with 8bit, it is fine. D3D9 with 10bit, it is fine.

Subsequently, D3D11 10bit output takes longer to go into exclusive fullscreen compared to D3D11 8bit or D3D9 10bit. Second(s) compared to instantaneously. If D3D9 can switch to fullscreen instantly with 10bit output, shouldn't D3D11 be able to do so just as fast?

Again, reproduction steps: Toggle fullscreen with D3D11 10bit, watch for queues to fill. Toggle back to windowed mode, watch for queues to fill. Toggle back to fullscreen, present queue does not fill and the video frames appear to stutter. Continuous playback the entire time, no pausing between toggling modes. Need to pause and unpause to get queues to fill again / get smooth playback.

No other software running in the background - only the media player. Changing my monitor to 24hz, the present queue starts at 6-7 from the get-go instead of filled to 14-15 like at 59.95hz. When trying to reproduce by toggling fullscreen, windowed, and fullscreen again, playback appears to be laggier but has the same dropped render queue and present queue with the higher refresh rate. (0-2 & 2-5 respectively) I usually just keep my monitor at 60hz, but figured I'd see if it is any different at 24hz for you. Unfortunately it isn't.

madVR 0.88.14

[aufkrawall]

Originally Posted by Anima123

Are you talking about Chroma or Luma here with SuperRes?

I mean this regarding luma, look at the medallion:

off:


on:


Screenshots were created before madshi asking about zoom.

[Eyldebrandt]

Originally Posted by madshi

For those of you still having some weekend hours left for playing, here's a pre-release version of the next madVR build:

http://madshi.net/madVR8814pre.zip

It has the new SuperRes version inside, plus a new chroma upscaler, and the AdaptiveSharpen algo. So some new things to play with. Official version will come soon.

First, the new SuperRes is no longer destructive as the old one in somes demanding scene with very high bitrate, which is a very good thing.
On others points, the result seems very similar to the old version with everything @ 0.0 except strength, which is also a very good thing to me, regarding it was my set up.
On aliasing and ringing control, the improvement seems really efficient.
SuperRes is definitely going on the right direction, regardless the material content.

About AdaptativeSharpen.
Obviously, it's by far superior to FineSharp and LumaSharpen, on each particularity they've got.
It's fast and the aliasing and ringing control is amazing regarding his strength of effect.
With moderate set, it's an all-around process for very HQ content.
But, on med sources, like 720p and DVD content, the result is ugly.
That's based on my personal taste, and i basically think that sharpen algorithms are negatives ways with low quality content.

So, again and again, madVR becomes better each time and that's a great state.

[JarrettH]

Originally Posted by madshi

Thank you. This is a good sample.

Those blue artifacts are caused by chroma ringing. They occur most strongly with super-xbr with AR disabled. Next one in the list if Lanczos4 without AR. Then later comes Lanczos3/4 with AR, because the AR doesn't remove all ringing.

However, already v0.88.13 removed the problem for super-xbr, because v0.88.13 has a very strict AR algorithm implemented for super-xbr chroma upscaling. So no blue artifacts in viewer's sample when using super-xbr for chroma upscaling with v0.88.13 (or newer), if AR is enabled.

@viewer you should test again when you can

[Eyldebrandt]

On 1080p content with 3440x1440 monitor.

Originally Posted by madshi

1) Which image upscaling/doubling algorithm do you like to use with SuperRes and why?

Super-xbr 100 or NNEDI3 32/64
Both are quite quivalent. NNEDI3 is more clean, with a better ringing control, Super-xbr is more natural and infinitely less expansive.

Originally Posted by madshi

2) Which values do you like for "strength" and "softness". Please note that the default values are strength=1.0, and softness=0.0. And these *may* be the best values. But you can still try other values to see if you like them more.

Strength is 0.35 and softness is 0.0, with HQ downscaling.

Originally Posted by madshi

3) The "use HQ downscaling" option changes the overall "look" of SuperRes a bit. Which look do you prefer? Please note that with the option turned on, you may have to increase the number of passes, because the SuperRes effect is slightly less intense with this option turned on.

With HQ downscaling, by far.

Originally Posted by madshi

4) How many passes should be used as default?

5 passes on most Blu-ray content.
With very soft pictures, i set it to 7.