madVR - high quality video renderer (GPU assisted)

I am sorry for off-topic, but could someone link me to a 12bit video or image sample? I want to compare 8bit to 12bit color depth, but I need a 12bit source for that.

[huhn]

and how are you going to output the image as 12 bit?

and just take the 16 bit gamma ramp from the 10 bit thread.

[Qaq]

Originally Posted by XMonarchY

I am sorry for off-topic, but could someone link me to a 12bit video or image sample? I want to compare 8bit to 12bit color depth, but I need a 12bit source for that.

There is no 12 bit output in madVR ATM:

Originally Posted by madshi

Windows does not offer any APIs to output 12bit. It supports 16bit output, but in a weird way and the GPU drivers don't handle that well. Because of that it's not supported by madVR.

@all
Any chance to get 12 bit API in Windows 10, DX 12?

[6233638]

Now that we have a 10-bit output, I tried using CalMAN with madTPG to create a LUT for my display, and I'm running into a few issues:

1. When double-clicking to exit FSE mode, the screen just turns black rather than returning to windowed mode.
2. madTPG doesn't work very well if you have a single display setup.

It's very awkward to switch between CalMAN to set up the reads and then quickly switch over to madTPG and make it fullscreen on the same monitor.
Ideally what would happen is that madTPG would run hidden in the background and then enter an always-on-top state or FSE mode as soon as it is being told to display patterns.
And once that series of patterns is completed, return to being hidden in the background.

Ideally there would also be a way to pause for 5 seconds at the initial switch to FSE mode, but I assume that's something which would have to be fixed on CalMAN's side of things.

[SecurityBunny]

MadVR 0.88.11

Workaround for queues not filling in D3D11 FSE mode does not work for me. I still have unfilled queues 1-4/20 render and 1-8/15 present (which should be 16?).

[madshi]

Originally Posted by jaju123

I have a green bar across the top with the AMD W10 drivers in windows 8.1 found here: http://forums.guru3d.com/showthread.php?t=399595

Happens for all W10 driver builds.
Everything is all the wrong color and weird too.

Weird. Other drivers work fine, I guess?

Originally Posted by cyberscott

I hope it won't cause any negative issues for others as well.

Hope v0.88.11 is working fine for you?

Originally Posted by InstantAli3n

Is this what's causing ReClock to act weird? Because since 0.88 came out ReClock will not detect framerate automatically anymore.

Or maybe it's just that ReClock is garbage because devs are too busy supporting bloatware. (Had to be said). I mean if it's so hard why not just have a "set to nearest major framerate" option.

FWIW, madVR is reporting the detected refresh rate via the madVR interface. If Reclock wanted, it could get the refresh rate from there. JRiver MC does exactly that for their VideoClock feature (which is similar to Reclock). But it seems Slysoft doesn't have a lot of interest in developing Reclock any further.

Originally Posted by ryrynz

with D3D 11 enabled present frame for every vsync on or off if I enter the madVR properties page whilst a file is playing back
the render queue drops to nothing and video stutters until the video is paused and played again Latest Nvidia driver on W8.1
This is with FSE enabled, no issues in windowed mode.

I suppose this is with a dual monitor setup? And you're playing the video on one monitor and then opening up the madVR settings dialog on the other monitor?

All the queues are fine as long as you don't open up the settings dialog?

Does it make a difference if the playback monitor is primary or not?

Originally Posted by MysteryX

Hey Madshi, you managed to get a lot of stuff going that isn't currently possible in AviSynth. I'm wondering, how different is it to program for a DirectShow Filter than for AviSynth?

For example, how difficult would it be to port SuperRes to AviSynth?

Or, would it be very difficult to plug madVR into an AviSynth filter? That may be ideal as it would give access to all madVR features.

No change for my to even try to port any madVR filters to AviSynth. However, it's not impossible to maybe someday consider making madVR itself available in some form to AviSynth. But not any time even remotely soon.

Originally Posted by James Freeman

Can anyone please suggest a proper shader to correct white balance?
I need to tweak R,G,B Levels (not Contrast, Brightness or Gamma) to correct LED projector extremely blue light source in software.

The hardware color correction is extremely poor and leaves an uneven colorful (red and blue) gradients, so it is unusable.

Not sure how such controls would work. Unless the 3 separate channels behave perfectly linear, just scaling or offsetting one channel might not produce the results you're expecting. That said, it would be relatively easy to add controls to allow you to offset and/or stretch R, G and B separately. But to be honest, nobody has ever asked about this before, so it doesn't seem to be needed by many users. So it's ultra low priority, unless a lot of people think it's suddenly super urgent.

Originally Posted by karamancho

whats the proper way to diagnose a problem?

the fullscreen exclusive mode wont play nice on my PC no matter the setting

in windowed mode everything works fine and I get around 30% GPU utilization.

in FSE mode I get

decoder queue 14-17/16
upload queue 5-8/8
render queue 5-8/8
present queue 1-8/8

and around 35% GPU utilization.

changing scaling algorithms doesn't change those numbers.

win 7 ultimate, nvidia geforce 210, driver ver. 341.44

how can I identify the problem?

Do the tips on this post help in any way? They're meant for DX11, but they might influence DX9, too:

http://forum.doom9.org/showthread.ph...88#post1724688

Originally Posted by XMonarchY

I am sorry for off-topic, but could someone link me to a 12bit video or image sample? I want to compare 8bit to 12bit color depth, but I need a 12bit source for that.

You can use the "smallramp.ytp" test pattern. It's 8bit, but if you upscale it with any of the madVR scaling algorithms (not DXVA!) to fill the whole screen, it will be more than 12bit. See first post of this thread, madTestPatternSource.

Originally Posted by 6233638

1. When double-clicking to exit FSE mode, the screen just turns black rather than returning to windowed mode.

Was this a one-time-only problem, or does it happen every time? Does this only happen with madTPG, or also with your media player?

Originally Posted by 6233638

2. madTPG doesn't work very well if you have a single display setup.

It's very awkward to switch between CalMAN to set up the reads and then quickly switch over to madTPG and make it fullscreen on the same monitor.
Ideally what would happen is that madTPG would run hidden in the background and then enter an always-on-top state or FSE mode as soon as it is being told to display patterns.
And once that series of patterns is completed, return to being hidden in the background.

Ideally there would also be a way to pause for 5 seconds at the initial switch to FSE mode, but I assume that's something which would have to be fixed on CalMAN's side of things.

I think the ideal approach would be to run madTPG on your HTPC, and then to use a laptop to run CalMAN. If you have to run both on the same PC, then yes, maybe that's currently not as smooth an experience as it could be. It's not something I can fix on my own, though. Would have to work with Calman together on that.

Originally Posted by SecurityBunny

MadVR 0.88.11

Workaround for queues not filling in D3D11 FSE mode does not work for me. I still have unfilled queues 1-4/20 render and 1-8/15 present (which should be 16?).

Have you tried this?

http://forum.doom9.org/showthread.ph...88#post1724688

[madshi]

madVR v0.88.11 released

http://madshi.net/madVR.zip

* fixed: SuperChromaRes produced green screen when downscaling video a lot
* fixed: at exactly 50% zoom, chroma was always repositioned using Bilinear
* fixed: some D3D11 objects were not properly released
* fixed: D3D11 FSE losing focus resulted in black screen
* fixed: one more D3D11 shutdown crash
* fixed: fixed one more OpenCL compilation problem
* fixed: crash when downscaling a lot, with media player asking for cropping
* fixed: SuperChromaRes produced artifacts with "use 10bit chroma/luma" buffer
* removed Shiandow's debanding algorith
* added workaround for one cause of queues not filling in D3D11 FSE mode
* modified "high" debanding preset one last (?) time
* SoftCubic skips anti-ring now (AR harmed SoftCubic more than it helped)

Based on your feedback, I've decided to stick to madVR's debanding algorithm, and to drop Shiandow's alternative algorithm. It was a close race, but in the end there were more users voting for madVR's algorithm, and it's also tried and proven, so I didn't see a good reason to drop my algorithm for Shiandow's, after all. Please note that the "high" preset now also analyzes gradient angles, which means it's slower than it used to be. If you want to get the "old" high preset back, simply activate the trade quality option "don't analyze gradient angles for debanding". That will give you exactly the same "high" deband as older madVR versions used.

FYI, I've setup two domains for madVR:

http://madVR.net
http://madVR.com

Both will point to the same homepage. For now it's just a one-page homepage, containing a brief description and some links. Maybe some day it will become a more complex homepage. The website "madVR.com" was maintained by a madVR user. He has kindly agreed to transfer the domain to me (thank you!), but the transfering could take a couple of days. So for now, to get to *my* madVR homepage, use "madVR.net".

[SecurityBunny]

Originally Posted by madshi

Have you tried this?

http://forum.doom9.org/showthread.ph...88#post1724688

Yep. The only way I can get the queues to fill is to reduce how many video frames shall be presented in advance to 6.

Downgrading to madVR 0.88.8, all queues fill perfectly fine. Any version newer than that, it doesn't. Isn't the workaround in 0.88.11 suppose to fix the issue and bring functionality back towards the latest build it worked? For what it's worth, using x64 version.

Oh damn, I feel so horrible that Shiandow's de-banding algorithm was pulled. Maybe in the future that algorithm will get much better? I shouldn't have criticized it....

madshi - thanks for the new madVR version! Should previous versions be uninstalled before replacing files with new version of madVR? Or maybe settings should be reset before replacing old madVR files with new ones? Are those proper instructions?

One more issue I have had with madVR for a VERY long time. Its about madTPG. Whenever I use dispcalGUI and madTPG, I set madTPG to "Use FullScreen" and "Stay on Top". Yet, at times when calibration starts, madTPG expands to FullScreen, but no patterns are shown. Then I press Esc or Alt+Tab and notice there is another madTPG window in the background that does display patterns. Now I always make sure there is only ONE madTPG process and only ONE mad* home cinema control process. I tried to toggle between "Disable OSD" enabling and disabling, but that made no different. Pretty much, I have to use madTPG in window mode, which is no big deal, but I don't understand why FullScreen mode creates this problem...

[nonowaimas]

Is there any way to install or use madVR without administrator rights?

[madshi]

feedback

-------

Now that we've summed up debanding, I'd love to hear your feedback about "image enhancements" and "upscaling refinement". That's a rather broad topic, covering FineSharp, LumaSharpen und SuperRes, both before and after scaling. Tons of options, so a lot to test, a lot to play with. Let me know what you think, what you like and what you don't like.

-------

Let me describe again the concept I had in mind for these algorithms: In my experience there are 2 different causes to get soft/blurry video playback:

A) Either the source is already blurry.

If this is the case, a good idea might be to sharpen the source before upscaling it any further. The reason for this is that sharpening algorithms work by analyzing the direct neighborhood of each pixel in the image. Upscaling causes the image to get bigger, which means the neighbor pixels in the original image are much further apart from each other in the upscaled image. But the sharpening algorithm only looks at the near neighborhood. Which means that if we sharpen after upscaling, although the sharpening still somewhat works, it loses much of its effectiveness. If the source is already soft to begin with, sharpening after upscaling won't do very much, so for soft sources my idea was to sharpen them before upscaling.

This is what the "madVR\processing\image enhancement" settings page is for. You can think of "softness" as being just another artifact of some sources, and we fix it by sharpening the source before we scale it. This only applies to soft sources, though, so my suggestion would be to not enable these settings by default, but only by demand, if you have a really soft source.

B) Even a sharp source becomes blurry if you upscale it a lot.

Doesn't matter which scaling algorithm you use. Even Lanczos8 and NNEDI3 become soft if you upscale an image a lot. The amount of softness introduced by upscaling is probably linear to the upscaling factor, I think. My hope here is that we can find a way to setup the "upscaling refinement" in such a way that every source maintains its natural sharpness after upscaling, regardless of which upscaling factor is used. I would like to find upscaling refinement settings which we can setup once and then forget, settings which work for any source and any upscaling factor. And I think the "secret" to make this work is by splitting the upscaling into multiple 200% upscaling steps. For each such step we apply a specific amount of sharpening, just so that the image afterwards has roughly the same sharpness as the source. If we do that, we can repeat the 200% upscaling steps multipe times, and still end up with the same sharpness the source had.

At least that's the idea. I think the option "refine the image after every ~2x upscaling step" should be the right one to use to realize the above approach. But for testing purposes I want to let you choose to only apply upscaling refinement once after all upscaling is done, so you can compare the effect this option has.

-------

Some specific questions I'd like to get an answer for:

1) In my tests medium "error upscaling quality" was quite a bit faster than high, while producing almost the same quality. Do you agree? Can I drop the "high" setting and simply always use "medium"?

2) It seems that running FineSharp and LumaSharpen *before* SuperRes has a much smaller effect than running them afterwards. So my question would be: Which in your opinion produces the better results: Running FineSharp / LumaSharpen before or after SuperRes?

3) Which combination of settings do you like most? SuperRes alone? Or in combination with FineSharp and/or LumaSharpen? Or maybe just FineSharp?

4) Does using the upscaling refinement options change your preferred upscaling/doubling algorithm? Which did you prefer without upscaling refinement? Which do you prefer with upscaling refinement?

5) Which SuperRes parameters do you like?

6) For experts: When upscaling with a large factor, which results in more than just one refinement pass (see debug OSD), do you prefer to refine the image after every ~2x upscaling step? Or do you prefer to refine the image only once after upscaling is complete? Of course when refining the image after every ~2x upscaling step, you should lower the values/passes to make the overall refinement strength more similar to refining only once.

7) For experts: Considering my goal (see above) to find a setting which works for every upscaling factor to maintain the sharpness of the original goal, which exact upscaling algorithm and which upscaling refinement options do you like? I'm hoping to end up with just a low/medium/high setting, where high would maintain the exact source sharpness, and where medium and low would be toned down settings do apply less strong sharpening. Low should provide a small but noticeable sharpness boost (compared to no refinement at all). Medium should be in the middle between Low and High.

P.S: Forgot:

8) Do you prefer "linear light" for FineSharp turned on or off?
9) Do you like the LumaSharpen "experimental limiter" turned on or off?

Thanks!!

[madshi]

Originally Posted by SecurityBunny

Yep. The only way I can get the queues to fill is to reduce how many video frames shall be presented in advance to 6.

Downgrading to madVR 0.88.8, all queues fill perfectly fine. Any version newer than that, it doesn't. Isn't the workaround in 0.88.11 suppose to fix the issue and bring functionality back towards the latest build it worked? For what it's worth, using x64 version.

Oh well. If you're willing to run some tests, I could make a couple of test builds to get to the bottom of this problem on your PC. You'd need to be available for tests, though, with a reasonable reply speed, for that to make sense. What do you think?

Originally Posted by XMonarchY

Oh damn, I feel so horrible that Shiandow's de-banding algorithm was pulled. Maybe in the future that algorithm will get much better? I shouldn't have criticized it....

Don't feel bad. I asked for feedback, and Shiandow wanted honst feedback, too. It would have helped nobody if you simply praised both algorithms. The feedback in this thread helped Shiandow to noticeably improve the quality of his algorithm. So I'm sure he doesn't regret the whole feedback and algorithm refinement cycle we did in this thread. His algorithm isn't "lost", it will be used in other renderers, just not in madVR (at least for now). FWIW, my decision was based on the feedback of several users, not just you alone.

Originally Posted by XMonarchY

madshi - thanks for the new madVR version! Should previous versions be uninstalled before replacing files with new version of madVR? Or maybe settings should be reset before replacing old madVR files with new ones? Are those proper instructions?

You can do that, but you don't have to. You can simply replace the files.

Originally Posted by XMonarchY

One more issue I have had with madVR for a VERY long time. Its about madTPG. Whenever I use dispcalGUI and madTPG, I set madTPG to "Use FullScreen" and "Stay on Top". Yet, at times when calibration starts, madTPG expands to FullScreen, but no patterns are shown. Then I press Esc or Alt+Tab and notice there is another madTPG window in the background that does display patterns. Now I always make sure there is only ONE madTPG process and only ONE mad* home cinema control process. I tried to toggle between "Disable OSD" enabling and disabling, but that made no different. Pretty much, I have to use madTPG in window mode, which is no big deal, but I don't understand why FullScreen mode creates this problem...

I think ArgyllCMS/dispcalGUI try to connect to the running madTPG instance, and if that fails they start a new instance themselves. So in your case it seems that sometimes ArgyllCMS/dispcalGUI seem to be unable to connect to the already running madTPG instance for some reason. I don't know why. It probably has nothing to do with FSE mode. I can't reproduce it on my PC. And seemingly the ArgyllCMS/dispcalGUI devs can't reproduce it, either, or they would have complained to me. So there probably isn't much we can do. You could try to narrow down the exact circumstances when this happens. Maybe that would help, or maybe not.

Originally Posted by nonowaimas

Is there any way to install or use madVR without administrator rights?

I think you do need at least to register the ax file using regsvr32.exe, to make most media players be able to use madVR. I'm not 100% sure if that's possible without admin rights, but I rather don't think so, because registering madVR means writing to registry keys which probably can't be written to without admin rights. There are some media players which can run madVR without installing it, though. I think JRiver MC does that. But I don't know if JRiver MC itself requires admin rights during installation or not...

[Vyral]

My feedback about "upscaling refinement" :
When upscaling from 720 to 1080 with Jinc 3AR, the image was a little bit blurry.
I've tested LumaSharpen and SuperRes but they were too strong for me.
I'm currently using FineSharp with a strength of 1.0 and linear light, everything else is default (mode=1 ; thinning=0.019 ; repair=0.25), and I'm very satisfied with it.

[madshi]

Originally Posted by Vyral

My feedback about "upscaling refinement" :
When upscaling from 720 to 1080 with Jinc 3AR, the image was a little bit blurry.
I've tested LumaSharpen and SuperRes but they were too strong for me.
I'm currently using FineSharp with a strength of 1.0 and linear light, everything else is default (mode=1 ; thinning=0.019 ; repair=0.25), and I'm very satisfied with it.

I like FineSharp a lot, too. But FWIW, LumaSharpen und SuperRes can be toned down, too, of course, e.g. by reducing the SuperRes passes, or all of the parameter numbers. It's a lot of values you can adjust, so it's kinda hard to "test it all". I know...

Of course 720 to 1080p is a rather small scaling factor (1.5x), so there the refinement options have a quite strong effect. If you choose a bigger upscaling factor (e.g. "Double Size" in MPC-HC), the sharpening by all the algorithms would be less strong. That's something I'll need to take into account, too, when defining the final Low/Medium/High settings.

[SecurityBunny]

Originally Posted by madshi

Oh well. If you're willing to run some tests, I could make a couple of test builds to get to the bottom of this problem on your PC. You'd need to be available for tests, though, with a reasonable reply speed, for that to make sense. What do you think?

Sure thing. I'm free.

[baii]

Since upscaling is gonna get mentioned, I need some help on setting profile with scale factor. Not sure if I did it wrong

if (scalingFactor.x > 1 ) "upscale"

else "normal"

This is my profile rule, but don't seem to always work? Am I missing something?

My initial impression of the refinements are that

the default strength (2.0) for fine sharp is lil bit high~
lumasharpen is cheap to use
superres for 720p+ source is pretty demanding.

[Anima123]

madshi,

Does upscaling refinement works for other upscaling algorithms, like Jinc or Bicubic? And how does it differs to image doubling algorithms, as NNEDI3 or NEDI?

[madshi]

Originally Posted by SecurityBunny

Sure thing. I'm free.

Great! Then please start with creating a debug log with 30 seconds of playback, using v0.88.11. Those 30 seconds should show how the queues are not filing. Please try to avoid switching back and forth between windowed <-> FSE all the time, to make the log easier to understand. Upload to e.g. tinyupload.com or similar site. Thanks!

Originally Posted by baii

This is my profile rule, but don't seem to always work? Am I missing something?

I don't know. Some more details would be helpful. Which video and target resolutions does it work with and which not? For each such combination please post the "movie resolution" and "target rectangle" information from the debug OSD (Ctrl+J). Just the text, no screenshots, please.

Originally Posted by Anima123

Does upscaling refinement works for other upscaling algorithms, like Jinc or Bicubic? And how does it differs to image doubling algorithms, as NNEDI3 or NEDI?

Yes, it even works for Bilinear upscaling.

Jinc, Bicubic, NNEDI3 and NEDI are all upscaling algorithms.

The upscaling refinement stuff instead is some sort of sharpening/enhancement, which is applied to the video frame after upscaling.

[foozoor]

This is what I use to improve/enhance my videos with vapoursynth.
After a lot of experimentations, this is the best for me for almost any videos.
I think it will be possible with madvr.

This is my post processing chain:

1) Uscale with needi3 (1,5x or 2x).
2) Sharpen with FineSharp with low strength (0,8 to 1,5).
3) Rescale the video to its original resolution, I use spline64.
4) Debanding.

[Anima123]

Originally Posted by madshi

Yes, it even works for Bilinear upscaling.

Jinc, Bicubic, NNEDI3 and NEDI are all upscaling algorithms.

The upscaling refinement stuff instead is some sort of sharpening/enhancement, which is applied to the video frame after upscaling.

When using SuperRes with Bicubic, for example, the SuperRes works directly on the targeted image size, right? While the image doubling algorithms might need to be downscaled or upscaled if the target size is not exactly 2x or 4x?

There's a lot of combinations to be tried and tested.