madVR - high quality video renderer (GPU assisted)

[iSunrise]

Originally Posted by cyberbeing

This doesn't work for me with Images, only Videos.

You´re right, jesus what a mess.

Originally Posted by Shiandow

PotPlayer can do it as well, although for some reason it isn't the default setting. You need to go to Preferences->Playback and then enable "load previous/next item in play folder".

Thanks a lot. However, that only seems to work for videos, too, like in the case of MPC-HC. Like cyberbeing said, it seems only Zoom Player can currently do both.

[MistahBonzai]

Originally Posted by iSunrise

MPC-HC seems to support what you want, you can use PgUp/PgDn for that, while PotPlayer seems to require you to load the files into the playlist directly, before you can browse through them.

I had already tried various key combinations in a vein attempt to get MPC-HC to jump to another image in the chosen folder. Pg Up/Pg Down displays the status message that either the 1st file in the folder or the last file in the folder is already loaded while the image stays put. I'm looking at play list possibilities right now but don't see how to configure multiple files... Keep me in mind if you should come across answers. And Yeah, the image display feature provides a superior means of monitoring adjustment effects over poking around in a video. That and I'm really liking how I can modify the images with the various filters :-)

Edit: I see you folks have been busy replying while I've been typing :-) I'll take a look at this "Zoom Player" thing... Thanks a bit later..$20 bucks! I'm sure it's worth it if ya need it but I haven't watched a movie in years. No thanks, I'll look elsewhere....

[secvensor]

Probably it is necessary fo me to reinstall OS.

[JonnyRedHed]

Actually I am experiencing an issue, a new one. As I watch the video full screen, at some random point the screen just goes black and the video continues to play as I can still hear the audio. If I click the screen it pauses and can restart if clicked again. But the screen remains black. I do not get a madvr crash reporter window, nor a zoom player crash window. So I have to ctrl+alt+del and go to task manager and close zplayer process.

The odd time the screen has gone white, not black.

It can happen within seconds of first playback, or 30 mins or more. The first time only occured after a few hours of viewing. A reboot does'nt seem to fix the issue.

using v0.87.6 and zoom player 8.7 b33

[MistahBonzai]

Originally Posted by secvensor

Probably it is necessary fo me to reinstall OS.

Have you tried to run "restore default settings.bat" in your MadVR folder as administrator?

1. If not then do it now.
2. If this doesn't appear to reset the madVR settings then run "uninstall.bat" as administrator.
3. Then delete the contents of your MadVR folder.
4. Then download madVr 87.6 (http://madshi.net/madVR.zip) and copy the contents to your old MadVR folder.
5. Then run "install.bat" located in that folder (as administrator).
6. Then configure madVR with very Conservative settings.
7. Load a simple (easy to run) video file (like 720p/23.976 or 24fps.
8. Better now? Congratulations!
9. No? - time to reload

[James Freeman]

madshi,
Thank You !!

0.87.6
* madTPG now forces ordered dither, but you can en/disable colored & dynamic

0.87.5
* madTPG now always uses monochromatic ordered dithering

Something changed? Whats the difference?

[QBhd]

Originally Posted by MistahBonzai

It's my understanding - and I might be wrong here - that if you are not actually utilizing an external calibration file (yCMS or 3DLUT) you don't "disable the GPU gamma ramps" - I have done extensive display level correction utilizing test patterns but have not used yCMS or 3DLUT.

Under "Color & Gamma" I don't "enable gamma processing" because I hadn't selected "disable the GPU gamma ramps" under "Calibration".

It seems to me that with both selected you may well be working one against the other..no?

I just wanted to say thanks... I am not sure when I started to disable GPU gamma ramps or where I even read that it might be a good idea... but I just watched a bunch of shows tonight and I think this solved my gamma woes. I did decide on setting my calibration for the TV to pure power curve of 2.5 (since that is what the EDID said the TV uses) and then set gamma correction to pure power curve of 2.35... and it's much more natural looking then leaving both options disabled... so I am guessing that the EDID is correct.

Thanks again

QB

[MistahBonzai]

Originally Posted by QBhd

I am not sure when I started to disable GPU gamma ramps or where I even read that it might be a good idea... Thanks again
QB

UR quite welcome I, like you, did the same thing. Although lotsa stuff has been written in the various 'expert guides' I had a hard time figuring it out. At one time I had them both selected - I could decrease one and increase the other and they operated backwards from what I expected

My problem stemmed from what I viewed as ambiguous wording pertaining to what was considered 'calibrated'. In the end I figured if I hadn't actually changed the files then I didn't want to disable them.

[Asmodian]

Originally Posted by MistahBonzai

I see you folks have been busy replying while I've been typing :-) I'll take a look at this "Zoom Player" thing... Thanks a bit later..$20 bucks! I'm sure it's worth it if ya need it but I haven't watched a movie in years. No thanks, I'll look elsewhere....

There is a Zoom Player Free edition, does it not include this feature?

[secvensor]

MistahBonzai
variant 9

[madshi]

Originally Posted by SamuriHL

Oh thank you thank you for the refresh fix for 8.1. Watch...now that you've worked around it, the stupid update 1 coming out in a month will break it again.

I hope not! So does my workaround work for you?

Originally Posted by Kalanoch

With the new builds (0.87.5 and 0.87.6), mpc-hc 1.7.3 (and 1.7.1) crashes immediately upon opening about 90% of the numerous video files I've tested it with. Sometimes it pops ups a "please wait a moment" dialog box (the madVR crash reporter I think, but it doesn't get a chance to finish) right before crashing. There doesn't appear to be any immediately noticeable pattern as to why some videos work but the vast majority do not. Some files will open and play fine once, but then closing and opening a second time results in a crash. Some will always work. At first I thought it was a xy-subfilter problem, but it will crash even on files that have no subtitles. If I revert back to 0.87.4, all of the problems go away.

Running Windows 7, nVidia GTX 670 with 327.23 drivers.

Edit: This is definitely an issue with windowed overlay. Disabling windowed overlay makes the new builds work correctly without crashing, but since my present times are terrible and I sometimes get frame drops if I don't use windowed overlay, that is not a workaround that I can use.

Oh man, that's bad. I think this is probably cause by the cosmetical improvements I made in 0.87.5. That seems to work fine for most users, but seemingly it crashes on your PC. Don't know why, though.

How do those crashes look like? Does MPC-HC automatically disappear without any trace left? Or is there still some sort of crash complaint window or something? And even if you wait a couple of seconds the madVR crash assistant doesn't appear?

Anyway, here's a test build:

http://madshi.net/madVRoverlayFix.rar

Does it fix the problem for you? If so, please also try resizing the window multiple times, just to make sure that's stable for you, too.

Originally Posted by seiyafan

More results: [...]

Interesting - thanks!

FWIW, I think it's probably not necessary to split "720 -> 1080p" and "720 -> 1440p", because both are doing 1x luma doubling. In the same way I think "1080 -> 1440p" will perform almost identical to "1080 -> 4K", because again it's only 1x luma doubling in both cases. However, if you do "720 -> 4K" with luma quadrupling, that would slow things down even more.

When using NNEDI3 luma doubling/quadrupling, the deciding factors for performance are only source resolution and neurons. The target resolution doesn't matter so much cause Bicubic up/downscaling is so cheap in comparison to NNEDI3. So basically you could do colums like "SD luma doubling", "SD luma quadrupling", "720p luma doubling", "720p luma quadrupling" and "1080p luma doubling". That would be quite interesting. I think those 5 columns would give us a rough performance idea about every possible scenario.

Of course for luma quadrupling it's hard to decide how many neurons to use for doubling and how many for quadrupling. The neurons for doubling are defnitely more important than those for quadrupling. But then doubling/quadrupling with 128/64 might look better than 256/16 - and perform faster, too. So I guess I would suggest something like 16/16, 32/16, 64/16, 128/32, 256/32.

Originally Posted by andybkma

Noob question time: Since we are on the subject of dithering with the recent mVR build, I noticed that LAV Video also has two dithering options, Ordered & Random with Random being the default. Since it seems that "Ordered Dithering" is new and is now the default option for mVR should we change LAV also to Ordered or leave it at the default of Random? Cheers

That would be a question for nevcairiel. He already said that some users didn't like ordered dithering because the dither pattern matched their display's internal dithering pattern. So that's why he changed the default to random dithering. However, I'm not sure which kind of dithering pattern LAV uses. Maybe with a better dithering pattern and with a dynamically changing dithering pattern on top of that ordered dithering could be default in LAV again, too. But I don't know if nevcairiel has any fun/time to work on dithering. After all it's not needed when using madVR, anyway. But really, that's for him to decide...

Originally Posted by James Freeman

Something changed? Whats the difference?

Graeme suggested that colored dithering might actually be good for calibration, or maybe not. So I decided to let the user decide whether to use colored or monochromatic dithering in madTPG.

[nevcairiel]

Originally Posted by madshi

That would be a question for nevcairiel. He already said that some users didn't like ordered dithering because the dither pattern matched their display's internal dithering pattern. So that's why he changed the default to random dithering. However, I'm not sure which kind of dithering pattern LAV uses. Maybe with a better dithering pattern and with a dynamically changing dithering pattern on top of that ordered dithering could be default in LAV again, too. But I don't know if nevcairiel has any fun/time to work on dithering. After all it's not needed when using madVR, anyway. But really, that's for him to decide...

LAV uses a default 8x8 bayer matrix with ordered dithering.
The biggest problem was in YUV->RGB conversion, because I used the same matrix for all three channels, it resulted in the pattern to be amplified - but I fixed that by shifting the pattern between channels now, and the amplified pattern effect went away (and random uses different random sources for all channels anyway)

I don't consider random dithering in LAV any worse then ordered dithering. When looking at the night&day difference between random and ordered in madVR (when targeting lower bitdepth), I wondered for a second if random dithering wasn't implemented slightly incorrectly that made it too strong.

In any case, since LAV will never have to do any dithering when you use it with madVR, this point is probably rather unimportant for this thread.

[huhn]

Originally Posted by QBhd

I just wanted to say thanks... I am not sure when I started to disable GPU gamma ramps or where I even read that it might be a good idea... but I just watched a bunch of shows tonight and I think this solved my gamma woes. I did decide on setting my calibration for the TV to pure power curve of 2.5 (since that is what the EDID said the TV uses) and then set gamma correction to pure power curve of 2.35... and it's much more natural looking then leaving both options disabled... so I am guessing that the EDID is correct.

Thanks again

QB

disabling the gpu gamma ramps simply disables, all calibrations so a 3d lut is not double changed. this should have an effect if you don't use a calibration.

if a calibrated screen is really important to you, you may should think about getting a colorimeter.

if you say madvr your gamma of you screen is 2.50 and then process gamma to 2.35 you should get a gamma of 2.35 if it's true that your display is gamma 2.5. your screen should be brighter now.

the best gamma should be choice n on your lightning condition in your room. typical 2.4 is what you want.

Originally Posted by MistahBonzai

UR quite welcome I, like you, did the same thing. Although lotsa stuff has been written in the various 'expert guides' I had a hard time figuring it out. At one time I had them both selected - I could decrease one and increase the other and they operated backwards from what I expected

My problem stemmed from what I viewed as ambiguous wording pertaining to what was considered 'calibrated'. In the end I figured if I hadn't actually changed the files then I didn't want to disable them.

see above setting gamma in "calibration" should't have an effect without "color & gamma" settings.

[seiyafan]

Thanks madshi! I overestimated the downscaling time that's why I separated the 720s into two columns, thanks for pointing it out.

Now, would someone lend me your 4k project for a week for testing?

[JonnyRedHed]

I'm still getting a black screen every now and again which all but freezes zoom player. No crash reporter pops up. Apart from this little issue all is good.

[seiyafan]

I am still not sure about how doubling/quadrupling works in one area. Suppose if I play a 480 video on 1440 monitor with only doubling enabled, it would use doubling to do 480 to 960 first, then upscaling from 960 to 1440?
And if quadrupling is enabled, it would go from 480 to 1960 then downscale to 1440? If that's the case when does it ever use doubling if quadrupling is enabled?

[nevcairiel]

Originally Posted by seiyafan

I am still not sure about how doubling/quadrupling works in one area. Suppose if I play a 480 video on 1440 monitor with only doubling enabled, it would use doubling to do 480 to 960 first, then upscaling from 960 to 1440?
And if quadrupling is enabled, it would go from 480 to 1960 then downscale to 1440? If that's the case when does it ever use doubling if quadrupling is enabled?

You can configure scale factors when which should be used.
When you play 480 at 1440, you can tell it to quadruple, or only double and scale the rest of the way, and when playing 720 at 1440, it would only need to double, not quadruple anymore.

[Q-the-STORM]

Originally Posted by seiyafan

I am still not sure about how doubling/quadrupling works in one area. Suppose if I play a 480 video on 1440 monitor with only doubling enabled, it would use doubling to do 480 to 960 first, then upscaling from 960 to 1440?
And if quadrupling is enabled, it would go from 480 to 1960 then downscale to 1440? If that's the case when does it ever use doubling if quadrupling is enabled?

If I remember correctly, doubling does one nnedi pass and quadrupling simply does another nnedi pass...

So yes, if you got doubling enabled, it would do nnedi 480 to 960 and upscale the rest to 1440...
If you got quadrupling if would do nnedi 480 to 960, then nnedi 960 to 1920, then downscale to 1440...

and if you play 720p, then it should only do doubling (one nnedi pass) 720 to 1440, even if you have quadrupling enabled...
it should only quadruple if the target resolution has not been reached with one nnedi pass...

[madshi]

Originally Posted by nevcairiel

LAV uses a default 8x8 bayer matrix with ordered dithering.
The biggest problem was in YUV->RGB conversion, because I used the same matrix for all three channels, it resulted in the pattern to be amplified - but I fixed that by shifting the pattern between channels now, and the amplified pattern effect went away (and random uses different random sources for all channels anyway)

I don't consider random dithering in LAV any worse then ordered dithering. When looking at the night&day difference between random and ordered in madVR (when targeting lower bitdepth), I wondered for a second if random dithering wasn't implemented slightly incorrectly that made it too strong.

An advanced 32x32 ordered dithering matrix looks significantly better than an 8x8 bayer matrix, IMHO. madVR's random dithering is similar to what is recommended by experts for audio processing. It's using TPDF, and the amount of noise is probably twice as high as you're using in LAV. To be honest, I'm not sure if the same principles apply in video processing as they do in audio processing. Maybe a simpler approach instead of TPDF would do, and a lower amount of noise. But I don't think I'll bother investigating that now, since 32x32 ordered dithering looks better to my eyes, anyway, and isn't any slower...

Originally Posted by JonnyRedHed

I'm still getting a black screen every now and again which all but freezes zoom player. No crash reporter pops up. Apart from this little issue all is good.

What does the task manager say about ZoomPlayer's CPU RAM consumption when it's frozen like that?

Originally Posted by seiyafan

Suppose if I play a 480 video on 1440 monitor with only doubling enabled, it would use doubling to do 480 to 960 first, then upscaling from 960 to 1440?
And if quadrupling is enabled, it would go from 480 to 1960 then downscale to 1440?

Yes, if that's what your configuration asks for.

[madshi]

For those of you guys who already have signs of withdrawal from the lack of dither testing, here's one more dither test build:

http://madshi.net/madVRlinearLightDithering.rar

The algorithms are basically the same as in v0.87.6, however, I've added Shiandow's linear light processing idea. The results look quite promising to me. At least switching between 8bit and 3bit doesn't brighten up the image, anymore. So it does exactly what it's supposed to do. Let me know what you think! FWIW, it will probably cost a bit of extra performance, but I hope it will not be too much. The test build always applies linear light processing for dithering, for both ordered dithering and the 2 error diffusion algorithms.