madVR - high quality video renderer (GPU assisted)

[madshi]

Originally Posted by kasper93

I'm happy From quick look it has really good balance. "low" change picture only a little, medium clear up image really nice and high get rid of all banding in my worst samples. Really impressive.

Good to hear.

Originally Posted by detmek

Works fine for me but, just to check - is it normal that High preset gives me lower rendering time compared to Low and Medium preset?

Yes, because "High" uses a simpler algorithm (no "angleBoost") by default.

Originally Posted by leeperry

Humm, it doesn't look nearly as striking on screenshots

Then you could upload a small sample where it's clear to see in motion?

Originally Posted by kasper93

madVRdeband10 crashes MPC-HC on close (didn't notice before), madVRdeband11-13 is crashing MPC-HC instantly after playback start.
https://dl.dropboxusercontent.com/u/16282309/madVR/madVR_dump_debug_build.7z (10 and 13)
https://dl.dropboxusercontent.com/u/16282309/madVR/madVR13_dump_release_build.7z (only 13)

You will get what's wrong

Before 10th version everything was working fine.

That's bad. Unfortunately the MPC-HC dumps don't help me much. If MPC-HC crashes that means its outside of madVR's code. So if madVR is at fault, it's doing something bad, but since the crash happens elsewhere I'm not sure which of my code is responsible. Are you sure that this really still happens with build 13? How can I reproduce the problem on my PC? MPC-HC doesn't crash on my PC with build 13. But then, I'm not using the latest MPC-HC build. Maybe that's the difference?

Anybody else having crashes with MPC-HC when using build 13?

Originally Posted by andybkma

Greets, how does this new mVR debanding feature effect those vids that are being realtime postprocessed with avisynth script sharpeners such as LimitedSharpenFaster (LSFMod) or FineSharp? Does the mVR debanding (low,med) smooth out at the end what has already been sharpened by the avisynth script?

The debanding tries to separate detail from banding. Banding is considered to be faint bands in gradients. Running a sharpener *before* debanding isn't optimal, but if it makes anything worse, it should be in the opposite direction than you fear: The sharpener might sharpen the banding artifacts, moving them over the threshold the debanding algorithm classifies as banding. So it's possible that sharpening before debanding could reduce the debanding smoothness. Anything more contrasty than faint bands should not be affected by debanding. So no, you shouln't lose any sharpness by running debanding after sharpening, at least if you stick to "low" or "medium" debanding strength.

Originally Posted by dansrfe

So switching to exclusive mode just abruptly stopped working for me and I've tried clearing all the settings, reinstalling madVR and MPC-HC but to no avail. Really no idea what's going on.

What does "abruptly" mean exactly in this context? Do you mean to say it works with the official build and stopped working with the debanding test builds? Or does this not have anything to do with the test builds at all?

Originally Posted by ryrynz

Seeing "Resetting Direct3d Device Failed 8876086c" messages using Latest MPC-BE SVN.
The whole screen goes black then after a couple of seconds I see the BE window with the above message or sometimes the video window is white.
Ive also seen the occasional green screen for a few seconds upon playback.
No issues with previous builds and I also can't reproduce any of these with the debug build.
I did however manage to get BE to crash, I took screenshots of all the tabs, not sure if any of this will help but hey. Definitely something up with this build.

http://www.sendspace.com/file/3050zl

Make sure you're really using test build 13. Those screenshots indicate that you're still using build 12. Can you reproduce all those problems with build 13? BTW, there's no need to make screenshots. See my reply to djfred93 1-2 pages back for more information how to easily get the bug report in text form.

Originally Posted by turbojet

I'm mostly happy about the medium preset but how to change avgDif to 2.0?

So you prefer avgDif to be set to 2.0 for the medium preset? Do you have a video sample where 2.0 works noticeably better than 1.8?

See build 10 release notes. They explain how to change custom settings.

Originally Posted by turbojet

Is don't analyze gradient angles the option that disables angleboost? It makes a difference in gpu usage but a lot less then old angleBoost did.

Yes. I've changed the angleBoost texture format in the hope that it might reduce GPU power consumption. It didn't help much for my GPU (HD4000), but maybe it helps with other/older GPUs.

Originally Posted by turbojet

Is there a way for 2.0/4.0/2.0/off to be used every time a file is opened?

Sure, just customize the medium preset, and then configure madVR to use medium by default.

Originally Posted by turbojet

Also a lot of stuff in registry that doesn't seem to fit any longer.

True. Just delete all those values. madVR will recreate the ones used now.

Originally Posted by turbojet

Think I figured it out, at HKCU\Software\madshi\madVR
[0]: Low preset
[1]: Medium
[2]: High

To customize edit them as Decimal not the default hexdecimal and set customOn to 1

It should show the default value and then +/- next to it. For example avgDif of Medium to 2.0 shows avgDif 1.8 +0.2 when using the deband hotkey.

Sure, you can customize using the registry. But isn't it much easier to use the comfortable way? See build 10 release notes.

[turbojet]

I don't currently have a sample where 2.0 looks better than 1.8, been using the latter with great results the past few weeks but I'll try 1.8 and keep an eye out.

I was trying the old keys, arrow keys make more sense. One thing I found is lightBoost can't be disabled in low and medium with arrows, minimum is 1.0. maxAngle only goes to 0.1 but does it have an effect when lightBoost is at 0?

[huhn]

[QUOTE=madshi;1652227
Thanks, another bug, embarassing. Hopefully everything's finally working now with this build?

http://madshi.net/madVRdeband13.rar[/QUOTE]

i run in a bug too in version 12 which is fixed with version 13 but the error port was broken too. nothing serious but there is a possibility that the bug report is broken.

[madshi]

Originally Posted by turbojet

I don't currently have a sample where 2.0 looks better than 1.8, been using the latter with great results the past few weeks but I'll try 1.8 and keep an eye out.

If you find samples where 2.0 looks noticeably better I'd be interested in hearing about that.

Originally Posted by turbojet

I was trying the old keys, arrow keys make more sense. One thing I found is lightBoost can't be disabled in low and medium with arrows, minimum is 1.0. maxAngle only goes to 0.1 but does it have an effect when lightBoost is at 0?

There is no "lightBoost", you mean "angleBoost"? If you configure "angleBoost" to 1.0 (which means no boost), madVR will disable the extra rendering steps, and you should save GPU performance.

Originally Posted by huhn

i run in a bug too in version 12 which is fixed with version 13 but the error port was broken too. nothing serious but there is a possibility that the bug report is broken.

I don't think so. At least I got a useful bug report from build 12 from one other user.

[ryrynz]

Originally Posted by madshi

Make sure you're really using test build 13. Those screenshots indicate that you're still using build 12. Can you reproduce all those problems with build 13?

Wasn't using 13, glossed over that post as I thought 12 was the latest. It's fine now.

Any chances of a deband on/off toggle?

Just want to run this one by you, one other filter I've often used in Avisynth is Psharpen, IMO it does a fantastic job of sharpening in a natural fashion.
Having come from sharpeners like LSFmod, Seesaw and Finesharp amongst many others over the years I've settled on Psharpen, would you take a look and see if you feel it's worthy for entry into MadVR?

I saw Fluxsmooth mentioned earlier, I used that for a long time but moved on to VagueDenoiser, if you're looking into denoising give it a shot.

[romulous]

madshi, I have a question, mainly on behalf of Zoom Player users. I notice that the default Zoom Player key mappings conflict with the way you set custom deband values. The arrow keys (all four of them) and F2 are all bound to particular Zoom Player functions, and when you try and set a custom value in madVR, the keys activate Zoom Player functions as soon as you press them. You can of course de-map all of those keys, set the custom values and then re-map those keys back to the original functions, but that is cumbersome. So, I was wondering if you could offer a way for Zoom Player to interact with the custom deband setting without having to de-map the hotkeys first?

Thanks!

[huhn]

Originally Posted by madshi

I don't think so. At least I got a useful bug report from build 12 from one other user.

i can try it again and make a screen with the error but if it works i make a "false" bug report and i don't like to do things like that.

i'm going to get an colorimeter but it is very hard to get information for that at the moment i like to go with the X-Rite EODIS3 i1Display Pro. my aim is to create 3d lut for madvr but is this even the right device or i'm on the wrong path with this ?

[michkrol]

Thanks for the new build. Everything works correctly here on MPC-HC, but I didn't have any problems with 'deband10' either.

Really like how the debanding works. As for the presets, they work as intended for me. Only high could be higher for some really bad content (bad source and encode), but then the detail loss is really visible, hence I hope the custom settings stay with us in final version.

Once again thanks for your hard work and thanks to everyone who contributed to the presets

[leeperry]

Originally Posted by madshi

Then you could upload a small sample where it's clear to see in motion?

Sure thing, I'll work on that and PM you a link ASAP.
BTW, I've been meaning to ask you for a while but would you consider keeping all the sub-sections in mVR's config panel permanently open? Like this:

It's great that it'll remember the last active section but the nesting is quite annoying when it all fits within the GUI anyway and that you are constantly forced to get your way through useless mouse clicks. I'm sure many ppl use mVR with a trackpad and/or from their couch so that would make this even more of an annoyance

Or for ppl with zillion displays, maybe you could simply remember what sections were open?

Hope you'll consider it,

[6233638]

I haven't had a chance to try out the presets in these latest builds yet, though the removal of NonGradientPenalty is somewhat concerning, as I found it to do a good job preserving details near black without artifacting as much as LocalContrast, or having as much of a performance hit.
I wish I had the time to spend going over everything in detail again, but things have just been so busy lately, that I've not had the free time to spend doing that sort of testing.

It's not specific to this build, but I've noticed that there are often "outlines" around objects when switching from fullscreen to windowed mode:


This is with debanding disabled, but I'm guessing it's a bug introduced by AngleBoost. They disappear after a few seconds of playback.

---

Something unrelated to that however, is that I was wondering about seeking performance.
I was having to seek through a couple of videos quite a lot the other day, which got me to change my settings to try and improve it.
To give you an idea of what was happening, it was seeking 10s at a time, with the button repeating every 0.1s

The way I had madVR set up before, seeking was very slow and uneven.
Switching to CPU decoding rather than GPU accelerated decoding helped, and of course I unchecked the "disable playback start until render queue is full" option.
After making those changes, it would seek smoothly, but then ever 6-8 seeks, it would freeze up for a second.

The only way I was able to get completely smooth seeking without it pausing was to use Fullscreen Exclusive mode, Windowed Overlay did not seem to help at all.

Is this simply due to the fact that you get better performance in FSE mode, or could it be something else to do with how frames are presented?
I've been avoiding FSE mode ever since installing Windows 8 last year, as madVR keeps switching my display to the wrong refresh rates (23/59 rather than 24/60) but maybe I'll just have to stick with it until that is resolved.

---

And now that I'm using FSE mode again, I'm getting serious problems switching between FSE and Windowed mode when there is a refresh rate change involved. With the maximum queue sizes, I get about 1fps displayed for the first 20s when switching to FSE mode. Reducing the queue sizes to 4 brings that down to about 5 seconds.

This only affects Fullscreen Exclusive and not Fullscreen Windowed mode, and only when there's a refresh rate change involved.

[kasper93]

Originally Posted by madshi

That's bad. Unfortunately the MPC-HC dumps don't help me much. If MPC-HC crashes that means its outside of madVR's code. So if madVR is at fault, it's doing something bad, but since the crash happens elsewhere I'm not sure which of my code is responsible. Are you sure that this really still happens with build 13? How can I reproduce the problem on my PC? MPC-HC doesn't crash on my PC with build 13. But then, I'm not using the latest MPC-HC build. Maybe that's the difference?

I've noticed that it doesn't happen with all files. Try with this one https://dl.dropboxusercontent.com/u/16282309/madVR/sample_002.mkv Doesn't matter which MPC-HC version you use.

It crashes always at the same point. Watch from the beginning its around 0:01

And crash is somewhere in "fading" code. When I disable debanding or set settings to (medium/medium) there is no crash. (medium/high) crash always... I think madVR crash internally and then when it is poked by mpc-hc everything crashes.

[dansrfe]

It basically stopped working after replacing with the latest deband build. But since then I've tried replacing it back with the latest official build, resetting to default settings, updating to the latest MPC-HC, updating to the latest stable nvidia drivers, and it's still the same problem. I'm not exactly sure what caused the problem which is why I don't know where to begin.

[sebdelsol]

Originally Posted by madshi

I'll think of something for (hopefully) the next official release. Stay tuned...

I've seen that you can change the rendering queue setting with IMadVRSettings, so I'll try that before pausing and calling repeatedly OsdRedrawFrame().
It may be good enough to prevent the GPU from going berserk in pause.

[Anima123]

I had crash of mpc-hc several times with test madVR 13 when I drag another video to the player while playing a video. It is easy for me to repeat the crashes.

[Ap3 n1nja]

Hello madshi,

I know that you are busy getting deband out officially, but it seems that sometimes playing a file that display subtitles partly out of movie resolution will cause madVR to crash, giving the exception message:

Access violation at address $4a4a3df1 in module 'madVR.ax'. Read of address $0.

The problem may stem from some interaction with or could be due to XySubFilter, since uninstalling the subtitle filter stops the crashing.

Here is a small sample file: http://www.sendspace.com/file/wdil85

You may need to reopen the sample file a couple times for it to crash.

Can anyone else reproduce this? I only tried it on PotPlayer.

[huhn]

@Ap3 n1nja

i stop after ~20 tests it works fine with xysubfilter and mpc hc 1.7.0.7754

[Ap3 n1nja]

@huhn

Thanks for testing it out! So, that means it might be due to PotPlayer. Or maybe even my video driver...

With MPC-HC, is it able to block XySubFilter with its internal filters list? For some reason, XySubFilter always connects to madVR even though I set XySubFilter's priority to block in PotPlayer's global filter manager.

[kasper93]

You can't block XySubFilter. madVR will load it even if you block it. Only way is to unregister it ;p

I can't reproduce issue with your sample. Are you sure you are using http://madshi.net/madVRdeband13.rar ? Anyway if madVR catch the crash just send report to madshi and he will fix it.

Mine crash is harder to trace...

[vivan]

Ap3 n1nja,
it's probably this issue http://code.google.com/p/xy-vsfilter.../detail?id=153

[sneaker_ger]

There's a test build up for XySubFilter which should include such old fixes:
https://dl.dropboxusercontent.com/u/...XySubFilter.7z