madVR - high quality video renderer (GPU assisted)

[Uoppi]

Hmm, I did try all possible "level" tags but surprisingly none of them had any effect on the picture. Adjusting "blacklevel" to -50 made a clear difference but I'm not at all sure if that's the same thing as switching "level".

Sure enough, the encode has just the black level wrong because other than being washed out (sort of like overlaid with a greyish veil or something), it's looking good and has decent bitrate too.

Anyway, I'd prefer tags as those keyboard commands are cumbersome when using a smartphone as the remote.

[QBhd]

Originally Posted by Uoppi

Hmm, I did try all possible "level" tags but surprisingly none of them had any effect on the picture. Adjusting "blacklevel" to -50 made a clear difference but I'm not at all sure if that's the same thing as switching "level".

Sure enough, the encode has just the black level wrong because other than being washed out (sort of like overlaid with a greyish veil or something), it's looking good and has decent bitrate too.

Anyway, I'd prefer tags as those keyboard commands are cumbersome when using a smartphone as the remote.

You can easily change the KB shortcut in madVR to something that is easier to work with.

QB

[ashlar42]

madshi, would you be so kind as to provide a full explanation about the video and audio clock calculations?

I know the refresh rate is displaying the refresh rate after deviation is taken into account, so if one knows the exact timings he used, he can get to the percentage of deviation. Is this right? How do you come up with the value for "display"? Why is it shown as an absolute value instead of a percentage, as it happens for the audio clock?

What about the audio clock deviation? Both audio and video are measured against the system clock, right?

I'm using a GTX 660 with HDMI audio onboard and I'm seeing very similar values in deviation (based on the above assumptions) for both audio and video. Meaning if I input exact timings for the resolutions/refreshes needed I see excellent results when bitstreaming (1 frame drop/repeat every x days or every 18 hours at the worst), which kinda makes sense according to your previous descriptions. If the two clocks (video and audio) deviate by a similar amount, playback is smooth.
I'd just like to understand the process fully.

Thanks so much for your help and for all the work you put in madVR.

[Uoppi]

Originally Posted by QBhd

You can easily change the KB shortcut in madVR to something that is easier to work with.

QB

Sure. Still I prefer not to use keyboard at all but have the most frequently needed Kodi functions displayed as easily accessible buttons on my smartphone remote (self-customized Unified Remote remote). And the remote's getting a bit crowded .

Really strange the "levels" tag has no effect whatsoever (none from these makes a difference: levels=PC|TV|fullrange|limited|doubleExp|tripleExp). Picture remains unbearably grey, unlike with "blacklevel" which does work.

EDIT: Using the level tag does change the level (verified by CTRL+ALT+shift+i) but the black level doesn't change at all at any of the settings.

[chros]

Originally Posted by Uoppi

Really strange the "levels" tag has no effect whatsoever (none from these makes a difference: levels=PC|TV|fullrange|limited|doubleExp|tripleExp). Picture remains unbearably grey, unlike with "blacklevel" which does work.

EDIT: Using the level tag does change the level (verified by CTRL+ALT+shift+i) but the black level doesn't change at all at any of the settings.

I just tried it for you: it's working as it should for me with v0.90.17 using filename: "foo levels=tripleExp.mkv"

[Uoppi]

Originally Posted by chros

I just tried it for you: it's working as it should for me with v0.90.17 using filename: "foo levels=tripleExp.mkv"

I'm on the latest madVR and Kodi DSplayer but no luck.

I have no idea if it should matter but anyway I even tried setting the whole chain to full RGB (I usually have GPU on full while madVR, Kodi & TV are on limited). OSD tells me levels are changed but picture looks the same. Which, like I said, seems very odd as the blacklevel tag does work. Could the "culprit" be Kodi, which has its own range setting?

[Stereodude]

Originally Posted by XTrojan

This 2.4 vs 2.2 battle is overall stupid.

Rec709 movies are supposed to be seen with 2.2 gamma.

If you want to raise the gamma anyway, enable "dynamic contrast" on your TV, just be sure it's originally calibrated to 2.2.

Unless ur TV is very old or a monitor calibrating to 2.4 has no point as modern TVs with dynamic contrast can handle "dynamic" higher gamma better than pure 2.4 gamma (atleast 2013+ models), basically less black/white crush.

It's not anywhere nearly as clear cut as you're suggesting. When they used CRTs for mastering when calibrated were not really delivering a 2.2 gamma. As such, you can't really claim they're supposed to be seen with 2.2 gamma. When mastering monitors were switched from CRTs to other display types I don't know.

[StinDaWg]

I have an odd but very specific bug to report. Using the latest AMD drivers (5/23 I believe), madVR v0.90.20, and HD7750.

When playing back 1080i content (either regular or film deinterlacing, doesn't matter), I get 1 line pixel flickering around all 4 borders, unless "use a separate device for presentation" is checked. This only affects D3D9 exclusive mode. Windowed and D3D11 are fine.

Edit: It's not just 1080i29.97 videos, it's also doing it on 720p29.97 videos.

Also, NNEDI3 isn't working at all. It just ignores the settings and uses whatever is set for image upscaling (Jinc, Lanczos, ect). Super-XBR works though.

[QBhd]

Originally Posted by StinDaWg

I have an odd but very specific bug to report. Using the latest AMD drivers (5/23 I believe) and HD7750.

When playing back 1080i content (either regular or film deinterlacing, doesn't matter), I get 1 line pixel flickering around all 4 borders, unless "use a separate device for presentation" is checked. This only affects D3D9 exclusive mode. Windowed and D3D11 are fine.

No help from me... but wow, that really is a very odd and specific bug :P

QB

[Uoppi]

Could someone else using Kodi DSplayer kindly check if the different "levels" tags make any difference to the picture (i.e. blacks)?

And as I'm out of luck with this, I'd like to know if finetuning blacks using the "blacklevel" tag is essentially the same thing as changing levels via the "level" tag? (Changing levels via the tv remote for every clip is too cumbersome and not a satisfactory option for me.)

[QBhd]

Originally Posted by Uoppi

Could someone else using Kodi DSplayer kindly check if the different "levels" tags make any difference to the picture (i.e. blacks)?

And as I'm out of luck with this, I'd like to know if finetuning blacks using the "blacklevel" tag is essentially the same thing as changing levels via the "level" tag? (Changing levels via the tv remote for every clip is too cumbersome and not a satisfactory option for me.)

It should just be an odd occurrence. If most of what you are watching is wrong levels, then maybe you have the wrong setup...

QB

[Uoppi]

Originally Posted by QBhd

It should just be an odd occurrence. If most of what you are watching is wrong levels, then maybe you have the wrong setup...

QB

Don't know what could be wrong with the setup. Black levels are calibrated OK and it's just this specific washed-out encode.

Anyway, shoud be easy to test with DSplayer and keyboard command CTRL + SHIFT + ALT + i. Wondering if it works for someone else.

[huhn]

Originally Posted by StinDaWg

I have an odd but very specific bug to report. Using the latest AMD drivers (5/23 I believe), madVR v0.90.20, and HD7750.

When playing back 1080i content (either regular or film deinterlacing, doesn't matter), I get 1 line pixel flickering around all 4 borders, unless "use a separate device for presentation" is checked. This only affects D3D9 exclusive mode. Windowed and D3D11 are fine.

do you have a sample?

[leeperry]

Originally Posted by Asmodian

Is the "cleanup image borders by cropping" option not a good solution? It is probably meant to clean up faded or distorted edges of analogue captures but it could work well here too?

It does but manually only so ideally I'd like to remux MKV music videos out of those big captured TS files and tag them accordingly in order to hide all borders/logos leftovers once and for all.

Anyway, my long delay when closing PotP or changing the frame rate in Reclock was due to some sloppy USB audio drivers from CEntrance and strangely it was still hiccuping when the DAC was disconnected, it really only stopped acting up after I uninstalled them.

Also, ever since I upgraded from mVR .19 to .20 I found the picture too sharp, I just did some A/B comparisons on very sharp and realistic looking 1080p30 content and as much as toggling the new deringer in .19 outrageously improves the picture as if it was cleaning fog from the camera optics .20 looks way oversharp and digital. I'll try to find some test patterns if need be, I'll stick to .19 for now.

[ashlar42]

Originally Posted by ashlar42

madshi, would you be so kind as to provide a full explanation about the video and audio clock calculations?

I know the refresh rate is displaying the refresh rate after deviation is taken into account, so if one knows the exact timings he used, he can get to the percentage of deviation. Is this right? How do you come up with the value for "display"? Why is it shown as an absolute value instead of a percentage, as it happens for the audio clock?

What about the audio clock deviation? Both audio and video are measured against the system clock, right?

I'm using a GTX 660 with HDMI audio onboard and I'm seeing very similar values in deviation (based on the above assumptions) for both audio and video. Meaning if I input exact timings for the resolutions/refreshes needed I see excellent results when bitstreaming (1 frame drop/repeat every x days or every 18 hours at the worst), which kinda makes sense according to your previous descriptions. If the two clocks (video and audio) deviate by a similar amount, playback is smooth.
I'd just like to understand the process fully.

Thanks so much for your help and for all the work you put in madVR.

Some more info. I'm trying, together with user hannes69 from Kodi forums, to understand exactly the process behind the numbers madVR displays.

Considering you have stated in the past that:

1) "display" shows the refresh rate adjusted for the video clock deviation measured against the system clock

2) "clock deviation" shows the deviation for the audio clock, again measured against the system clock.

I'll make an example of what we found for a 59.940 video content played back with bitstreaming audio at 59.940 display refresh rate.

Display: 59.94174
Clock Deviation: +0.00276%
1 frame drop every 3.17 hours
movie 59.940 fps

audio clock deviation 27.6ppm
video clock deviation: (59.94174 - 59.94) / 59.94 = 29.03ppm
total deviation: video clock dev - audio clock dev = 28.86ppm - 27.6ppm = 1.43ppm

frames in 3.17 hours = (3600 * 59.940) * 3.17 = 684,035.28

1 frame dropped every 684,035.28 frames. Accuracy of 1 / 684,035 = 0.0000014619, that is 1.46ppm (I guess this is well within the margin of error in madVR approximations once it reaches the x.xx hours stage).

The above tells me that you are using the approximation of 59.940 in the internal calculations, and not the 60/1.001 value. Is this correct?
If using the 60/1.001 value we would have:

audio clock deviation 27.6ppm
video clock deviation: (59.94174 - (60/1.001)) / (60/1.001) = 28.03ppm
total deviation: video clock dev - audio clock dev = 28.03ppm - 27.6ppm = 0.43ppm

frames in 3.17 hours = (3600 * (60/1.001)) * 3.17 = 684,035.96

1 frame dropped every 684,036 frames. Accuracy of 1 / 684,036 = 0.0000014619, that is again 1.46ppm

0.43ppm clock difference audio/video seems too low if compared to what drop/repeats at every 3.17 hours indicate, while 1.43 is very similar.
So, are you using 59.940 for internal calculations? Does this hold true for 23.976 as well?

Sorry, I know we are reaching the human limits of geekiness here... but we would really love to fully understand the underlying math

I also can't understand why my clock deviation sits constantly at about 0.00276% at 59.940 refresh rate while going down to 0.00242% at 23.976 refresh rate. It's the audio clock measured against the system clock. Why should display refresh rate influence the deviation? I have been checking and checking but the difference between the two refresh rates stays there... It's audio and video through HDMI on a GTX660, if it matters.

[Werewolfy]

Originally Posted by madshi

madVR v0.90.20 released

http://madshi.net/madVR.zip

* tweaked deringing algorithm some more
* added workaround for weird crash with latest AMD beta drivers

Not much time for madVR this weekend, so only small changes. Feedback about the tweaked deringing algo welcome.

Thanks for the new version, the tweaked deringing algo is definitely better. Much less artifacts and deringing is still very powerful. Ok sometimes it removes some details but a lot less that others deringing algo I've seen so far so it doesn't bother me, I know it won't never be perfect.

The only real artifact that remains is the one I showed you... Do you need a sample of this video?

But I admit I find that the pros of deringing outperform the cons that remain this time

[leeperry]

Houston, we have a problem(that's with .19 BTW):



Seems boiling down to FS(whatever E or W) + SSIM 2D + black bars detection as either going 1D or disabling the latter works fine, here's a MKV sample.

So I captured a big .TS file off DVB-T2 and only this music video in the sample gave that first problem but once I enable doubling they all go bananas like this in FS:



PS: Oh and I wish I could get black bars to never ever appear not even for a few secs as I'm not watching a movie but simply rolling music videos in big .TS captured files, am I missing an option here?

[huhn]

i get the green screen with always - 2x supersampling image doubling + SSIM 2D target resolution 1920x1080p.
crop black boarder doesn't have an effect on this for me.

this doesn't happen with "always - if upscale is needed" only with "always - 2x supersampling". i guess this has something to do with the 1440x1080p 16/9 source that is displayed at 1920x1080p.

i'm doing this in mpc-hc window mode so the resolution is the problem not FS. i'm on UHD.

i'm using madVR 90.20 and 960 gtx 368.22

[baii]

Originally Posted by StinDaWg

I have an odd but very specific bug to report. Using the latest AMD drivers (5/23 I believe), madVR v0.90.20, and HD7750.

When playing back 1080i content (either regular or film deinterlacing, doesn't matter), I get 1 line pixel flickering around all 4 borders, unless "use a separate device for presentation" is checked. This only affects D3D9 exclusive mode. Windowed and D3D11 are fine.

Edit: It's not just 1080i29.97 videos, it's also doing it on 720p29.97 videos.

Also, NNEDI3 isn't working at all. It just ignores the settings and uses whatever is set for image upscaling (Jinc, Lanczos, ect). Super-XBR works though.

The driver probably broke opencl. I haven't bothered to upgrade to any of the crimson AMD driver , after testing 5-6 versions for almost half year and none of them can just "work".

[MariaX9]

After some Days of testing, i am now using this Settings.

Processing
- artifact removal: medium/high,
- reduce ringing artifacts, reduce dark halos
everything other is deafult

Sacling algorithms
- Chroma: super-xbr 100 AR, SR 2
- Downscaling: SSIM 2D 100% AR, LL, AB 75%
- Doubling: luma/chroma always 2x, super-xbr AB25
- Upscaling: Jinc, AR
- Refinement: sharpen edges 0.5, crispen edges 1.0, AB 100%, AR, SR 2

I usually watch 480p and 720p Content. The only settings im not sure about, are the Downscaling AB strength and all the Settings at Refinement. May someone has any suggestions?