madVR - high quality video renderer (GPU assisted)

[leeperry]

I get a black screen with .14, .13 is fine. Something to do with that pesky W7SP1 platform update I can't apply again I guess? :o

I'll provide a log if obie's doesn't help.

[resides]

Heads up!

Black screen here too.

[har3inger]

Crispen edges (and the old FineSharp) suffer a lot in enhancing source aliasing.
Posterized (cel shading) images get dramatically changed by thin edges set at high values, but this is probably by design.

So far so good. Enjoying the new algos.

[Warner306]

Is it possible to improve the performance of Sharpen Edges for fair testing? It is far too expensive on my system.

[ryrynz]

I get a black screen with .14, .13 is fine. Something to do with that pesky W7SP1 platform update I can't apply again I guess? :o

I'll provide a log if obie's doesn't help.

Getting that here too. Found the issue.. DXVA upscaling/downscaling.

[cvrkuth]

I get a black screen in madvr after I updated to 0.89.14 from 0.89.13. I did not change any parameters, I just unpacked the new zip over the old install.
+1
Same here

[ryrynz]

+1
Same here

Is it DXVA upscaling/downscaling related for you too?

[cvrkuth]

Is it DXVA upscaling/downscaling related for you too?

No, I use default settings for upscaling/downscaling, not DXVA ...

[pjred]

Me too, black screen 0.89.14 even with default settings.

[Magik Mark]

Madshi,

All features are running perfectly. Only draw back is it added addition 10ms on rendering time from my default of around 12ms. I'm now getting around 22 to 23ms

[wanezhiling]

The black screen issue of version 0.89.14 seems to be related to the video decoder, at least it happens with mpc-be and PotPlayer built-in software decoder here.

[FreeFall]

madshi,

Here is a new sample, the same original DVD sample I uploaded but has a re-cap of events in the first movie that plays before starting. I cut it to reduce the size of the original but is seems to be the cause of the problem.

https://www.dropbox.com/s/6ifrmre5xlbtnmk/testdisc.zip?dl=0

Zoom control settings used.

move subtitles, to bottom of the screen / window.
automatically detect hard coded black bars.
notify media player about about cropped black bars, no more than once every 2 sec.
keep black bars visible if they contain subtitles, forever.

The black bars are being cropped, it seems to be working as intended in ver 0.89.9. I tried disabling the move subtitles option but it had no effect.

Thanks.

[krille]

Would it be possible to ad gpu and maybe cpu load to the madvr stats window?

[nlnl]

Quote:
Originally Posted by nlnl View Post
The problem: When source is video (29,97) and "force video mode' option is on, display does not switch to 59 rate, but swtches to 23.
If I set to 'video mode' using keyboard shortcut display switches.

Hmmmm... Which decoder? Software decoding? DXVA copyback? Native DXVA? Does this problem occur with all videos or just some?

LAV 0.66.0.27, DXVA copyback, the problem occurs with all videos (29.97 fr), not films.
Some times it switches when playing video first time but does not switch when playing the second.

[madshi]

Crispen edges (and the old FineSharp) suffer a lot in enhancing source aliasing.
Do you have a sample where this is especially obvious? I doubt there's much I can do about it, but I'd at least like to have a look at it.

Is it possible to improve the performance of Sharpen Edges for fair testing? It is far too expensive on my system.
As I said a couple posts before, performance of the new algos is not fully optimized yet. Performance optimizations costs time and a lot of thought. So I wanted to first get feedback about the quality (not performance) of these new algos. Maybe some algos are bad and should be deleted instead of optimizing them? Want to get confirmation that the algos produce nice quality and are worth spending the time to optimize them on.

All features are running perfectly. Only draw back is it added addition 10ms on rendering time from my default of around 12ms. I'm now getting around 22 to 23ms
I hope the slowdown only happens when activating the new sharpening algos? madVR itself didn't get slower, right?

Here is a new sample, the same original DVD sample I uploaded but has a re-cap of events in the first movie that plays before starting. I cut it to reduce the size of the original but is seems to be the cause of the problem.

https://www.dropbox.com/s/6ifrmre5xlbtnmk/testdisc.zip?dl=0

Zoom control settings used.

move subtitles, to bottom of the screen / window.
automatically detect hard coded black bars.
notify media player about about cropped black bars, no more than once every 2 sec.
keep black bars visible if they contain subtitles, forever.

The black bars are being cropped, it seems to be working as intended in ver 0.89.9. I tried disabling the move subtitles option but it had no effect.
Are you using the current official LAV version, or a newer nightly build? Newer nightly builds now send the subtitles to madVR instead of burning them into the decoded image. This has the advantage of madVR getting full control over the subtitle drawing and positioning.

When using a newer nightly LAV build, madVR should be able to move the subtitles either into the black bars or into the active video area. However, these are bitmap subtitles, which means that for optimal quality madVR has to draw them onto the decoded video image before doing any scaling, because only this way the subtitles will get the full quality benefit of the scaling algorithm you've selected. Having to draw the subs before scaling means that if you tell madVR to actually crop the black bars away, there's no black bars left for madVR to draw the subtitles into. So if you want the subs to be drawn into the black bars, you have to deactivate the option "crop black bars" in the zoom control settings. (Please note that this only applies to bitmap subtitles.)

Would it be possible to ad gpu and maybe cpu load to the madvr stats window?
Many things are possible, but at this point I don't consider that important enough to spend my time on. Too many other things on my to do list which I consider more important.

LAV 0.66.0.27, DXVA copyback, the problem occurs with all videos (29.97 fr), not films.
Some times it switches when playing video first time but does not switch when playing the second.
Can you please try if anything changes if you use software decoding? Also, can I have a debug log from a failure situation (I don't need one from a success situation)? Thanks.

[madshi]

madVR v0.89.15 released

http://madshi.net/madVR.zip

* fixed: black screen with some decoders (introduced in v0.89.14)
* fixed: black screen with DXVA processing in some situations
* added "superResRadius" file name tag for testing purposes

[KoD]

Bug report: AR video is broken in 0.89.14 and 0.89.15. Works ok with older versions. Tested this on 16:9 video with ass subs, which is for whatever reason displayed as 4:3 video (and the subs are offset badly). Sample here (http://s000.tinyupload.com/index.php?file_id=43695178824048275421).

[madshi]

Bug report: AR video is broken in 0.89.14 and 0.89.15. Works ok with older versions. Tested this on 16:9 video with ass subs, which is for whatever reason displayed as 4:3 video (and the subs are offset badly). Sample here (http://s000.tinyupload.com/index.php?file_id=43695178824048275421).
Are you using a newer MPC-HC nightly build? And do you have "keep aspect ratio" enabled? If so, please try whether unchecking "keep aspect ratio" fixes the problem. Updating to the latest MPC-HC nightly build should also fix the problem - if it is what I think it is.

[KoD]

It happens with both latest two MPC-HC nightly builds (190 and 202). Having "keep aspect ratio" checked or not does not matter. The sample I provided is only 500Kb, so it's easy to download it and test.

Edit: looks like madVR ignores the container set AR (which is 16:9) and instead uses the video stream AR (which is 4:3).

[nlnl]

Can you please try if anything changes if you use software decoding? Also, can I have a debug log from a failure situation (I don't need one from a success situation)? Thanks.
Can not reproduce using software decoding.
Debug log:
https://drive.google.com/file/d/0B7a6LffuxvKURTN3UnRjTktzalU/view?usp=sharing]

[seiyafan]

There will be little to no difference unless you are using SuperRes. SuperRes is more effective when used with Image Doubling.



I've never seen a real difference between Jinc AR or image doubling for 1080p -> WQHD, neither with SuperRes. I'm watching every 1080p video upscaled to WQHD.



Thanks for confirming that, I was thinking about the same thing.

[Thunderbolt8]

is anyone able to tell the difference of what sharpen, crisp, thin edges are doing?

and what exactly "enhance detail" means? is it normal sharpening? or do just some things get sharpened and others dont? on what basis?

[ibius]

How about you read at least 1-2 pages back first?
Madshi's second post one page back covers that.

[madshi]

It happens with both latest two MPC-HC nightly builds (190 and 202). Having "keep aspect ratio" checked or not does not matter. The sample I provided is only 500Kb, so it's easy to download it and test.

Edit: looks like madVR ignores the container set AR (which is 16:9) and instead uses the video stream AR (which is 4:3).
I had tried and wasn't able to reproduce the problem. But I tried again and now found that the problem only occurs with "detect black bars" disabled. Should be fixed in this test build. Can you confirm?

http://madshi.net/madVR8915arFix.rar

Can not reproduce using software decoding.
Debug log:
https://drive.google.com/file/d/0B7a6LffuxvKURTN3UnRjTktzalU/view?usp=sharing]
Ah, ok, I understand the problem.

In theory the stream could switch between progressive and interlaced all the time, which would mean madVR would have to switch deinterlacing on and off all the time. If I changed refresh rate every time that happens that could be very annoying. So atm madVR only changes refresh rate in very specific situations: 1) When playback starts, if the settings say so. 2) When going fullscreen, if the settings say so. 3) When manually activating video mode deinterlacing.

Not sure whether there are real world videos out there which often switch back and forth between progressive and interlaced?

Software decoding is not affected because it's fast enough to switch madVR into video mode deinterlacing. DXVA takes too long, so at the moment when madVR wants to switch refresh rates, DXVA has not yet told madVR that the video is interlaced.

[KoD]

I had tried and wasn't able to reproduce the problem. But I tried again and now found that the problem only occurs with "detect black bars" disabled. Should be fixed in this test build. Can you confirm?

http://madshi.net/madVR8915arFix.rar
The issue is fixed with this build.

[madshi]

The issue is fixed with this build.
Ok, will release a new official later today, will take a couple of hours, though.

Could you please bring back the radius option for a build or two?
You can add e.g. "superResRadius=80" to the file name now to force a SuperRes radius of 80.

[AngelGraves13]

Did some testing with SuperRes and will still leave it off as it sharpens the image undesirably. It didn't look too good when I was watching 720p content at 1440p. It was too sharp. Disabled and all was back to normal. I guess SuperRes isn't for me.

[MS-DOS]

AngelGraves13, if you tested it with one of the latest two builds then maybe it's because the Radius option was removed and forced to 0.66, which makes it look way too sharp and aliased sometimes. I used it with 1.00 before and it was totally fine to me.

[aufkrawall]

You could try a higher radius like madshi explained above or a lower strength.

I noticed that after playing around with different upscaling refinements, madVR somehow disabled NNEDI3 and used Jinc instead. It couldn't be activated again unless madVR was restarted. Haven't been able to reproduce yet, but there must be a small problem.

[AngelGraves13]

AngelGraves13, if you tested it with one of the latest two builds then maybe it's because the Radius option was removed and forced to 0.66, which makes it look way too sharp and aliased sometimes. I used it with 1.00 before and it was totally fine to me.

I'm mostly watching 1080p blu-rays from my personal collection that I've remuxed to avoid using discs. The content looks fine with Jinc AR for Chroma and Image. SuperRes doesn't make it look better.

[mzso]

There are a bunch of sharpening algorithms now. Is there some writing about it? How do they compare to each other quality wise. Or in what situation do they come handy?
For a long time I used the players sharpencomplex2 shader. Now I started using madVR's adaptivesharpen, though to be honest I don't know if it's the best one (for me).

[aufkrawall]

I'm mostly watching 1080p blu-rays from my personal collection that I've remuxed to avoid using discs. The content looks fine with Jinc AR for Chroma and Image. SuperRes doesn't make it look better.

imho fonts or emblems look too blurry with 1080p -> WQHD without SR. Especially true for fine white fonts.

[Sunset1982]

I'm using madvr to upscale 1080p Bulray remux files to a 4k Display but there is a lot of noise. Any suggested settings to get rid of it?

[pankov]

Honestly, do you have any videos which need rotation *and* have subtitles?? I thought about subtitles, but it's pretty hard to implement correctly, and my thinking was that videos that need rotation are very unlikely to have subtitles. So I simply didn't spend any time on working on rotated subtitles. If you have legit files which need rotation and have subtitles, please let me know where they come from and why they are rotated. Then I'll think about whether it's worth spending time on...

To be hones no, I don't have such video files - it was just an imaginary use case and since you say that it's not easy to be implemented I totally agree with your decision not to invest time in it until there is a legit need for it.

[Thunderbolt8]

How about you read at least 1-2 pages back first?
Madshi's second post one page back covers that.I found the post you are refering to. just dont refer to page numbers please, some people have different posts per page settings.

[6ari8]

Thanks for the new build madshi!

I'm an AdaptiveSharpen user and I must say I'm really happy with the new sharpen edges option. I'm torn between the two but I think I'll settle on one of them after some tests.

However, there may be a bug in the new build's "upscaling refinement" options.
When choosing SuperRes on its own, the "refine the image after every...."/"refine the image only once...." is greyed out even though it takes effect according to the checked option of the two.
So If I wanted to change that option I have to check one of the other refinement options temporarily so it becomes visible.

[obieobieobie]

Another black screen log for 0.89.15 with unchanged settings from 0.89.13 where there is no black screen with this clip:

fryingpanfail.webm

log here (dunno why the log is HUGE):
http://s000.tinyupload.com/index.php?file_id=11448852747541458473

[XMonarchY]

Damn, I still get stutters after pausing and then resume videos and nothing I've tried or that was recommended made any difference... Its SO annoying.

[djsolidsnake86]

the aspect ration problem finally is gone

the only problem now is the stuttering when going fullscreen with a video with mpchc

[madshi]

I noticed that after playing around with different upscaling refinements, madVR somehow disabled NNEDI3 and used Jinc instead. It couldn't be activated again unless madVR was restarted. Haven't been able to reproduce yet, but there must be a small problem.
If you find a way to reliably reproduce it, let me know.

There are a bunch of sharpening algorithms now. Is there some writing about it? How do they compare to each other quality wise. Or in what situation do they come handy?
For a long time I used the players sharpencomplex2 shader. Now I started using madVR's adaptivesharpen, though to be honest I don't know if it's the best one (for me).
Asked and answered twice already in the last couple of pages.

I'm using madvr to upscale 1080p Bulray remux files to a 4k Display but there is a lot of noise. Any suggested settings to get rid of it?
There's no denoise algo in madVR. You could try ffdshow raw video processing with some AviSynth noise reduction script. Sorry, can't tell you any more than this. That's a topic for a different thread. In this thread all I can say is: madVR currently doesn't do denoising.

To be hones no, I don't have such video files - it was just an imaginary use case and since you say that it's not easy to be implemented I totally agree with your decision not to invest time in it until there is a legit need for it.
Ah ok.

I'm an AdaptiveSharpen user and I must say I'm really happy with the new sharpen edges option. I'm torn between the two but I think I'll settle on one of them after some tests.
Let me know what you settle on and for what reasons. Please judge quality and performance separately.

However, there may be a bug in the new build's "upscaling refinement" options.
When choosing SuperRes on its own, the "refine the image after every...."/"refine the image only once...." is greyed out even though it takes effect according to the checked option of the two.
So If I wanted to change that option I have to check one of the other refinement options temporarily so it becomes visible.
In the next build SuperRes will always only be applied once, because it simply doesn't make sense any other way. SuperRes has no benefit from being applied multiple times.

Another black screen log for 0.89.15 with unchanged settings from 0.89.13 where there is no black screen with this clip
I think this should be fixed in v0.89.16.

Damn, I still get stutters after pausing and then resume videos and nothing I've tried or that was recommended made any difference... Its SO annoying.
the aspect ration problem finally is gone

the only problem now is the stuttering when going fullscreen with a video with mpchc
Please create a debug log with 20 seconds worth of stuttering. Afterwards please stop MPC-HC by pressing Alt+F4, then zip and upload the debug log.

Can not reproduce using software decoding.
Debug log:
https://drive.google.com/file/d/0B7a6LffuxvKURTN3UnRjTktzalU/view?usp=sharing]
I've done a change in v0.89.16 which should hopefully fix this problem.

[madshi]

madVR v0.89.16 released

http://madshi.net/madVR.zip

* fixed: occasional aspect ratio problems (introduced in v0.89.14)
* fixed: "refine only once" doesn't affect SuperRes, anymore
* fixed: DXVA decoding + deint -> refresh rate sometimes didn't switch

[Warner306]

As an ardent SuperRes user, I haven't spent any time testing the new sharpening algorithms with upscaled content.

Using Image Enhancements at 1080p -> 1080p:

Sharpen Edges: Too expensive to test, but promising based on a couple of samples.
Crispen Edges: The edge enhancement is noticeable, even at low settings, but the image lacks a uniform increase in detail, making the image appear flat. Not really a fan.
Thin Edges: At 1080p, this degrades the image and is not of any use.
Enhance Detail: For some reason, this appears to be more of a direct replacement for FineSharp. Edge enhancement is not obvious. While the image appears sharper overall. Noise is likely amplified but is not noticeable with the sources I'm using.

I have barely used each algorithm, but I'm having a hard time finding a favourite algorithm. I haven't found an equivalent setting to match the result of SuperRes. SuperRes does not crease the corners of objects as much as these sharpeners, which is why I currently prefer Enhance Detail.

[Asmodian]

The issue with anamorphic video I noticed with "automatically detect hard coded black bars" disabled has been fixed with v.0.89.16. Thanks!

I am getting a huge performance hit when using NNEDI3 quadrupling and any image enhancement set to "refine the image after every ~2x upscaling step".
e.g.
NNEDI3 quadrupling and AdaptiveSharpen with "refine the image only once after upscaling is complete", 28.5 ms
NNEDI3 quadrupling and AdaptiveSharpen with "refine the image after every ~2x upscaling step", 54.4 ms
super-xbr quadrupling and AdaptiveSharpen with "refine the image only once after upscaling is complete", 16.3 ms
super-xbr quadrupling and AdaptiveSharpen with "refine the image after every ~2x upscaling step", 17.2 ms

GPU clock speed was constant throughout this testing.

Is this large of a difference expected? I understand there is extra work when moving in and out of YCbCr for NNEDI3 but this is larger than expected. The hit from the NNEDI3 64 quadrupling itself is less than that from running AdaptiveSharpen with "refine the image after every ~2x upscaling step". :eek:

edit: I should mention that this doesn't bother me as super-xbr quadrupling is excellent, and I can use more upscaling refinements with it, but I thought it was odd so I should mention it. :)

[madshi]

Crispen Edges: The edge enhancement is noticeable, even at low settings, but the image lacks a uniform increase in detail, making the image appear flat. Not really a fan.
Enhance Detail: For some reason, this appears to be more of a direct replacement for FineSharp. Edge enhancement is not obvious. While the image appears sharper overall. Noise is likely amplified but is not noticeable with the sources I'm using.
FWIW, these two do exactly the opposite thing. The first just works on edges but not "flat" areas, while the second just works on flat areas but not edges. This way you can fine tune how much "enhancement" you want to apply to edges and textures separately. The "crispen edges" algo intentionally only looks at edges and nothing else, so that you can fine tune every aspect of the sharpening process to your exact likings.

These algos are not "either this or that", you can combine all 4 of these to achieve the exact result you want.

Thin Edges: At 1080p, this degrades the image and is not of any use.
It depends on the source. Some sources have very fat lines, for these "thin edges" can be great. But mostly it's useful after upscaling. Try a tiny bit of "thin edges" after SuperRes, you might like it.

[madshi]

I am getting a huge performance hit when using NNEDI3 quadrupling and any image enhancement set to "refine the image after every ~2x upscaling step".
e.g.
NNEDI3 quadrupling and AdaptiveSharpen with "refine the image only once after upscaling is complete", 28.5 ms
NNEDI3 quadrupling and AdaptiveSharpen with "refine the image after every ~2x upscaling step", 54.4 ms
super-xbr quadrupling and AdaptiveSharpen with "refine the image only once after upscaling is complete", 16.3 ms
super-xbr quadrupling and AdaptiveSharpen with "refine the image after every ~2x upscaling step", 17.2 ms

GPU clock speed was constant throughout this testing.

Is this large of a difference expected? I understand there is extra work when moving in and out of YCbCr for NNEDI3 but this is larger than expected. The hit from the NNEDI3 64 quadrupling itself is less than that from running AdaptiveSharpen with "refine the image after every ~2x upscaling step". :eek:
I haven't really done benchmark tests for these things, but obviously running those algos multiples times costs performance. My future plan is to try to modify/tune the sharpening algos to only have to run once after upscaling, but still achieve a good effect, regardless of the upscaling factor. Currently if the upscaling factor is very large, most of the sharpeners (except SuperRes) lose their effectiveness.

[nlnl]

madshi
* fixed: DXVA decoding + deint -> refresh rate sometimes didn't switch
Thanks!
Sorry, but the problem is not fixed here :(
The log:
https://drive.google.com/file/d/0B7a6LffuxvKUSS1UZTRyeGx6Ylk/view?usp=sharing

[baii]

The issue with anamorphic video I noticed with "automatically detect hard coded black bars" disabled has been fixed with v.0.89.16. Thanks!

I am getting a huge performance hit when using NNEDI3 quadrupling and any image enhancement set to "refine the image after every ~2x upscaling step".
e.g.
NNEDI3 quadrupling and AdaptiveSharpen with "refine the image only once after upscaling is complete", 28.5 ms
NNEDI3 quadrupling and AdaptiveSharpen with "refine the image after every ~2x upscaling step", 54.4 ms
super-xbr quadrupling and AdaptiveSharpen with "refine the image only once after upscaling is complete", 16.3 ms
super-xbr quadrupling and AdaptiveSharpen with "refine the image after every ~2x upscaling step", 17.2 ms

GPU clock speed was constant throughout this testing.

Is this large of a difference expected? I understand there is extra work when moving in and out of YCbCr for NNEDI3 but this is larger than expected. The hit from the NNEDI3 64 quadrupling itself is less than that from running AdaptiveSharpen with "refine the image after every ~2x upscaling step". :eek:

edit: I should mention that this doesn't bother me as super-xbr quadrupling is excellent, and I can use more upscaling refinements with it, but I thought it was odd so I should mention it. :)
What is the initial video size? I recently went up to 4k and any refinement got more expensive as it is done @4k. If you use every 2x, would it be possible that those refinement was done at high res then downscale to wqhd( was it you with wqhd and 980ti iirc?)

Sent from my 306SH

[obieobieobie]

I think this should be fixed in v0.89.16.

Thank you madshi, I can confirm it is fixed.

[madshi]

Sorry, but the problem is not fixed here :(
The log:
https://drive.google.com/file/d/0B7a6LffuxvKUSS1UZTRyeGx6Ylk/view?usp=sharing
Ok, will look at it next week end.

[Asmodian]

I haven't really done benchmark tests for these things, but obviously running those algos multiples times costs performance. My future plan is to try to modify/tune the sharpening algos to only have to run once after upscaling, but still achieve a good effect, regardless of the upscaling factor. Currently if the upscaling factor is very large, most of the sharpeners (except SuperRes) lose their effectiveness.

But why would it cost so much more when run twice with NNEDI3 compared to super-xbr? AdaptiveSharpen is running the same number of times and at the same resolutions in both cases but the performance hit is 2600% greater when using NNEDI3 quadrupling. I am using NNEDI3 256 doubling in both cases.

I agree, AdaptiveSharpen is more effective when run twice in this situation. Looking forward to the tweaks for large scaling factors. :)

What is the initial video size? I recently went up to 4k and any refinement got more expensive as it is done @4k. If you use every 2x, would it be possible that those refinement was done at high res then downscale to wqhd( was it you with wqhd and 980ti iirc?)

Sent from my 306SH

The source res was 720x480 (anamorphic 16:9). However notice the difference between super-xbr and NNEDI3. I get only one extra ms from running AdaptiveSharpen twice (once at 1706x960 and once at 3412x1920 instead of only once at 3412x1920) using super-xbr quadrupling but I get an extra 26ms when running AdaptiveSharpen twice using NNEDI3 quadrupling.

Running AdaptiveSharpen only at the quadrupled res of 3412x1920 costs ~1.4ms using either NNEDI3 or super-xbr quadrupling.

Yes, I am running WQHD with a 980Ti. :)