madVR - high quality video renderer (GPU assisted)

[Warner306]

Image enhancement is applied before scaling, upscaling refinement is applied after upscaling. So yes, if you activate both, and then upscale, both will be applied separately ("twice"). At least that's the case atm. I have some ideas how to maybe streamline the whole thing, but the algos will need improvements before I can do that.

A good way to streamline image enhancements and upscaling refinement options is to combine the two into a category called "Image Sharpening" or "Post-process Sharpening" and separate them into "Pre-resize" and "Post-resize." One category to rule them all.

[QBhd]

I have a quick question... Is upscaling refinement applied after a downscale?

QB

[har3inger]

I have a pretty general question about scaling which I think people in this thread are likely well positioned to answer, but apologies if this goes too far off topic. Basically, I'm wondering if you have an HD source which itself has been upscaled from SD but using a poor algorithm (e.g. bilinear), is there likely to be any benefit in downscaling it back to SD and re-upscaling it using a better algorithm? Or would it always look worse?

I'm pretty sure yes, but you'd need to use a fancy algorithm like deblinear to undo the upscale. Otherwise, you introduce artifacts when you upscale with the bilinear, then introduce even more when you downscale with something that isn't its exact inverse. Oh, and you'd need to know the source resolution of the master before it was upscaled poorly.

[huhn]

I have a quick question... Is upscaling refinement applied after a downscale?

QB

i did a short test and no didn't worked.

i don't know if sharpening is better done after downscaling or not but i guess it is.

[leeperry]

I think you didn't understand me: Running SuperRes every 2x upscaling step simply doesn't do anything. It looks pretty much the same as running it only once after upscaling is complete. So it's a total waste of performance to run it after every 2x upscaling step. As I said, if you think I'm wrong then please provide screenshot proof, using an older madVR build (please keep AdaptiveSharpen/FineSharp/LumaSharpen disabled for those screenshots).
Oh, so we're back to the "pretty much"/"more or less" argument :(

"screenshot proof"? I don't think there is such as thing as a "ground truth" concept when it comes to subjective settings such as sharpening? Otherwise there would be no strength option to those four new features to begin with?

What would change your mind exactly? Using a 1080p reference, downscaling it to anything using 960x540 using CR AR LL and then showing differences with 2X SR checked?

I just did a whole bunch of A/B comparisons on DVD'ish-resolution noisy movies in .13 with sxbr25+SR(St4/Sh2/R0.66) and I far prefer having 2X SR checked because it does look like double SR filtering to me, I mean I do get double-SR and high motion looks less edgy and a whole lot smoother(30fps@60Hz with Reclock) when the option is checked.

You admitted that 2X SR might make sense when other sharpening refinements are checked and it would make total sense to me to have SR pick the best looking frames after every 2X upscale, I think an extra checkbox as to whether the end-user wants SR going 2X as well would make perfect sense.

NNEDI3 is the real GPU hog, my HD7850 doesn't break a sweat when 2X SR is enabled and either way the PQ improvement totally warrants the extra GPU load(which is very low anyway).

I'm far more convinced by 2X SR than the four new features tbh because SR is just much more technically advanced to my eyes, ensuring that only the best frames are selected instead of trying to polish a turd so to speak.

I'm not interested in what "sharpen edges" and "crispen edges" have to offer, "thin edges" might be worth looking into and I'm not sure about "enhancing details" yet but then again I've never been interested in regular sharpening filters, I only like SR because it's a lot smarter than a regular sharpness knob and it looks outstanding in motion.

Long story short 2X is totally worth the extra GPU load to my eyes, at least with Reclock on a LCD screen and "thin edges" might allow sxbr to look thin like NNEDI3.

:thanks:

[Shiandow]

I'm far more convinced by 2X SR than the four new features tbh because SR is just much more technically advanced to my eyes, ensuring that only the best frames are selected instead of trying to polish a turd so to speak.


You seem to be misunderstanding something, SR doesn't pick anything, it just works with the one frame. Or, to use your terminology, it's just really good at polishing turds.

[madshi]

A good way to streamline image enhancements and upscaling refinement options is to combine the two into a category called "Image Sharpening" or "Post-process Sharpening" and separate them into "Pre-resize" and "Post-resize." One category to rule them all.
I have a better plan (I think), but as I said, it won't work without doing some improvement to the algos first.

I have a quick question... Is upscaling refinement applied after a downscale?
No.

Oh, so we're back to the "pretty much"/"more or less" argument :(

"screenshot proof"? I don't think there is such as thing as a "ground truth" concept when it comes to subjective settings such as sharpening? Otherwise there would be no strength option to those four new features to begin with?
We're back at "leeperry argues and argues and fails to provide any screenshots", as usual. This is my 3rd reply to your request, and you still didn't provide any screenshots. You're wasting my time by debating and debating and not doing what I asked for already in the my first reply to you, namely providing some screenshots.

All I want is some screenshots that show that running SuperRes after every 2x upscaling looks better than running it only once. That's all. No ground truth needed.

[ace960]

When next file in a directory is played then the audio is played only. The video freezes at the previous file. I mean, I have several files in single directory. I use mpc-hc with option to play next file in the directory after playback of the actual file ends. When the next file should be automatically played then the audio starts correctly. However, the video freezes at the last frame of the previously ended file. The same happens when I skip to the next file manually (video freezes at the last played frame of the previous file). I found out that this started with madvr 0.89.13. It is present in all subsequent releases 0.89.14-16. The problem is not present in 0.89.12. I started having the problem while running mpc-hc 1.7.9.190. Now I use mpc-hc 1.7.9.202 and it makes no change. I also use reclock 1.8.8.5. All software is 32-bit, running on Windows 7 64-bit (running geforce drivers 358.91, problem was seen also on 358.50).

[leeperry]

You seem to be misunderstanding something, SR doesn't pick anything, it just works with the one frame. Or, to use your terminology, it's just really good at polishing turds.
Yes exactly, 2X does a more efficient double filtering and yields less noise in the end providing smoother motion thanks to the cleaner edges. Sharpening noise doesn't work.

We're back at "leeperry argues and argues and fails to provide any screenshots", as usual. This is my 3rd reply to your request, and you still didn't provide any screenshots. You're wasting my time by debating and debating and not doing what I asked for already in the my first reply to you, namely providing some screenshots.

All I want is some screenshots that show that running SuperRes after every 2x upscaling looks better than running it only once. That's all. No ground truth needed.
Last time I spent forever providing screenshots you dismissed them so I'm asking about a procedure that would make sense to you for a change.

We are indeed back at the "it looks more or less the same" argument on your part but again at this rate a lot of options in mVR are redundant, especially when it comes to chroma.

2X SR doesn't look "pretty much the same" by a long shot with noisy SD on my rig and you agreed that it made sense when extra refinements options were enabled so I don't understand why the sudden decision to disable it :(

So if I enable 0.1 "thin edges", I get 2X SR? I guess I could live with that.

And we're also back at the argument that I'm big enough to run scripts on my side(lucky me I don't have to justify my VST plugins chain in ffdshow to anyone) and that I'll have to waste time proving my point again when the difference is *very* obvious in motion on my rig and that this option had been there since the beginning for a good reason. Oh well, as a ground truth isn't needed this time I'll try to find noisy material that shows cleaner edges with 2X SR.

:thanks:

[chros]

don't change the gamma to fix crushed blacks. use the custom levels for it. way more effective and it "doesn't" change the gamma of the picture
Thanks for the suggestion, huhn, I'll try it out.
But what can be the difference at the viewer's point? I mean we already modified the gamma on the TV to 2.4.

In my case, on my TV I find 2.2 too bright overall, and 2.4 (even though it comes out of black very quickly, shadow areas are far too bright with both 2.2 and 2.4) a bit too dark overall and a bit too contrasty. Setting the TV to 2.2 and having MadVR apply a 2.35 gamma curve seems just about perfect.
That's pretty interesting. I thought it should be almost the same with: TV 2.4 and mdvr default gamma (2.2).

[huhn]

Thanks for the suggestion, huhn, I'll try it out.
But what can be the difference at the viewer's point? I mean we already modified the gamma on the TV to 2.4.

this just fixed crushed blacks. without really touching the gamma curve.

why do a double gamma correction that shouldn't even fix issue like that?

and don't even assume you get the gamma you setup in a TV. without calibration and a meter you don't know what you get and you can assume you have more of an S curve gamma than anything else.

for a dark room you should aim at 2.4 anyway.

[Murmeltier]

I have a Samsung S24D390HL display which has a AD-PSL. The panel is not able to support 8-bit colors natively. It does 6-bit + FRC from what I read.

What am I supposed to choose under "Device->*->Properties->The native display bitdepth is? 6-bit I guess? If this is the case. Isn't there a conflict between madVR's dithering (which obviously is better) and that FRC dithering the monitor does automatically? I also guess I'm not able to deactivate the latter?

[nevcairiel]

I have a Samsung S24D390HL display which has a AD-PSL. The panel is not able to support 8-bit colors natively. It does 6-bit + FRC from what I read.

What am I supposed to choose under "Device->*->Properties->The native display bitdepth is? 6-bit I guess? If this is the case. Isn't there a conflict between madVR's dithering (which obviously is better) and that FRC dithering the monitor does automatically? I also guess I'm not able to deactivate the latter?

Pick whatever looks better to you. It depends on the quality of the dithering algorithm in the screen, really.

[jkauff]

I have an HP Stream 7 tablet that I very seldom use. This weekend, though, I'll be watching a couple of movies on it. The processor is an Atom Z3775G quad-core, and the unit runs 32-bit Win 8.1. I can run MPC-HC using QuickSync for decoding, and madVR works at the default settings (looks pretty good, too). Being a tweaker, though, I won't be satisfied until I see how far I can push this chip.

Assuming I can only make a few settings changes in madVR until I start dropping frames, what changes are likely to give me the most improvement with the least cost? The screen is 1280 x 800 and the movies I'll be watching are SD. Doubling is probably out of the question, so what else should I try to tweak?

[mogli]

With some old interlaced DVDs I now have some repeated frames (no drops) while using smooth motion since madVR recently introduced automatic switching to DXVA scaling for both chroma and luma when deinterlacing. Does this mean it's a NVIDIA driver bug or still a madVR one?

[har3inger]

I have an HP Stream 7 tablet that I very seldom use. This weekend, though, I'll be watching a couple of movies on it. The processor is an Atom Z3775G quad-core, and the unit runs 32-bit Win 8.1. I can run MPC-HC using QuickSync for decoding, and madVR works at the default settings (looks pretty good, too). Being a tweaker, though, I won't be satisfied until I see how far I can push this chip.

Assuming I can only make a few settings changes in madVR until I start dropping frames, what changes are likely to give me the most improvement with the least cost? The screen is 1280 x 800 and the movies I'll be watching are SD. Doubling is probably out of the question, so what else should I try to tweak?

I think DXVA2 is better than quicksync. Use that instead.

Smooth motion is really good, and is probably going to give you the biggest boost if your monitor doesn't support 24fps. The jump from SD->1280x800 isn't too big, so jinc AR for luma is probably sufficient and not too different from what doubling would produce. I don't think you can run SuperRes, but if you can, that's the next thing to try turning on. Use bicubic75 AR or jinc3 AR for chroma if you can afford it.

Checking on 10 bit image/chroma buffer in trade performance for quality is unlikely to result in any visible difference. Those would be worth using for you as well. Disable linear light for dithering is also fine, but I'd leave linear light on for smooth motion because it makes a big difference there.

Ordered dithering is great, and you should use it since it's better than all the alternatives for you.

Debanding can help immensely for sources with banding (obviously), but can cost a bit. Depends if you think your videos need it.

[leeperry]

All I want is some screenshots that show that running SuperRes after every 2x upscaling looks better than running it only once. That's all. No ground truth needed.
Your wish is my command: http://screenshotcomparison.com/comparison/150162

All pictures as a .zip file (https://mega.nz/#!fhBlxaqY!fv2e7e_bTuMTG9QXNvCgAXlwjxWChkQ_8puMZT2R4Bg), using .13 with sxbr25+GL SR(St4/Sh2/R0.66), no other refinement was checked and I didn't try LL. Dithering was disabled.
The cleaner edges provide smoother motion, kinda dreaming of 4X SR now tbh and SuperSampling too while we're at it http://forum-images.hardware.fr/images/perso/sniperlk.gif

New GPU's will be out soon and I'd happily use them for outrageous SS SR.

[mcn]

Hi all,

I'm experiencing a strange v-sync issue with madVR.

In a two monitors setup if I have one or more applications open that use madVR after a random amount of time I start experiencing loss of v-sync in other applications.
To restore v-sync I have to minimize or close all the instances of the applications that use madVR.
Then, if I reopen some application that use madVR, after some time it happens again.

I clearly experience image tearing with applications like VMware Workstation, Outlook 2010 and VLC.
Strangely applications that use madVR aren't affected.

Below more details.

OS
The most important thing that I observed is that this started to happen just after upgrading a few weeks ago from Windows 8 N to Windows 8.1 N, both x64.
On Windows 8 I had installed the Media Feature Pack but on Windows 8.1 I hoped to do without it so it isn't installed.

GPU
I have a GTX 680.
I currently am on driver 358.50, but I tried various previous versions with no luck.
I changed some settings in the Nvidia control panel but this didn't help either.

Monitors
Initially I had one monitor connected via DisplayPort and the other via DVI (digital dual link).
Now they are both on DVI but it doesn't help.

Both monitors are 1920x1200 @ 60 Hz.
Nvidia control panel shows 60 Hz, Windows says 59 Hz and madVR something like 59.95 Hz.
From what I understand this is normal.

If I use only one monitor there are no v-sync issues.

It doesn't seem to matter on which monitor the applications are shown.

Software
I'm pretty sure the issue is triggered by madVR because it happens if I have open instances of DVBViewer Pro and/or MPC-HC, both 32 bit and using madVR.
Also I tried VMR 7 and 9 with apparently no issues.
I can't try EVR because I'm on an N edition of Windows.
I'm on madVR 0.89.16.
Again I tried different settings and versions of madVR without finding a solution.

I also tried using another video decoder instead of LAV Video (32-bit, v0.66) but the issue remains so I really think that this is triggered by madVR.

I looked at madVR's OSD both when the issue is present and when it isn't and I couldn't see any relevant difference in stats to get a hint about what's causing the issue.



At this point I really think the only things that I could try are to install the Media Feature Pack for Windows 8.1 or get another GPU.
I'm not too fond of the first option because I'd like to keep my system as lean as possible.
And I'm quite satisfied with my GTX 680.

If anybody has any suggestion or wild guesses it would be really appreciated, thanks.

[jkauff]

I think DXVA2 is better than quicksync. Use that instead.

Smooth motion is really good, and is probably going to give you the biggest boost if your monitor doesn't support 24fps. The jump from SD->1280x800 isn't too big, so jinc AR for luma is probably sufficient and not too different from what doubling would produce. I don't think you can run SuperRes, but if you can, that's the next thing to try turning on. Use bicubic75 AR or jinc3 AR for chroma if you can afford it.

Checking on 10 bit image/chroma buffer in trade performance for quality is unlikely to result in any visible difference. Those would be worth using for you as well. Disable linear light for dithering is also fine, but I'd leave linear light on for smooth motion because it makes a big difference there.

Ordered dithering is great, and you should use it since it's better than all the alternatives for you.

Debanding can help immensely for sources with banding (obviously), but can cost a bit. Depends if you think your videos need it.
Tried your suggestions, and the only setting that pushed the GPU over the edge was jinc for luma. Had to stick with lanczos. Smooth motion On seems to produce the most visible improvement.

Thanks for the advice.

[huhn]

@mcn

try overlay mode may fix the issue.

[onyx76]

You think this combination is better than SuperRes? I haven't tried it with upscaled sources.

After several tests with upscaled sources I found that the IQ is better in my setup with this settings for all type of content:

Chroma: Jinc + AR
Image: Jinc + AR
Double Luma: 1.2x or greater - NNEDI3 64 Neurons
Quadruple Luma: 2.4x or greater - NNEDI3 64 Neurons
Downscaling: Catmull-Rom + AR + LL
Upscaling Refinement: Sharpen Edges (0.5) + Enhance Detail (0.5) + SuperRes (4 +LL)
Artifact Removal - Debanding: Low/High
Image Enhancements: Off
Dithering: ED2 + Use Colored Noise

Render time: 27ms.

[ryrynz]

Madshi, will image enhancements be shown on the OSD? Or perhaps set an option to allow them to be shown?

Image thinning produces very similar results to awarpsharp but I still prefer the lines it makes (tends to bow things out a bit less) is it possible for the line thinning can be tweaked to produce perhaps similar results? Maybe even something the user could set?

Oh and if you got a chance to add line darkening to the toolbox that's be great.

[madshi]

When next file in a directory is played then the audio is played only. The video freezes at the previous file. I mean, I have several files in single directory. I use mpc-hc with option to play next file in the directory after playback of the actual file ends. When the next file should be automatically played then the audio starts correctly. However, the video freezes at the last frame of the previously ended file. The same happens when I skip to the next file manually (video freezes at the last played frame of the previous file). I found out that this started with madvr 0.89.13. It is present in all subsequent releases 0.89.14-16. The problem is not present in 0.89.12. I started having the problem while running mpc-hc 1.7.9.190. Now I use mpc-hc 1.7.9.202 and it makes no change. I also use reclock 1.8.8.5. All software is 32-bit, running on Windows 7 64-bit (running geforce drivers 358.91, problem was seen also on 358.50).
Does the media player still react to mouse and keyboard events? E.g. can you press the stop button and it reacts?

With some old interlaced DVDs I now have some repeated frames (no drops) while using smooth motion since madVR recently introduced automatic switching to DXVA scaling for both chroma and luma when deinterlacing. Does this mean it's a NVIDIA driver bug or still a madVR one?
That's a cosmetical OSD bug. Will be fixed in the next build. You can safely ignore this, it has no meaning.

Your wish is my command: http://screenshotcomparison.com/comparison/150162

All pictures as a .zip file (https://mega.nz/#!fhBlxaqY!fv2e7e_bTuMTG9QXNvCgAXlwjxWChkQ_8puMZT2R4Bg), using .13 with sxbr25+GL SR(St4/Sh2/R0.66), no other refinement was checked and I didn't try LL. Dithering was disabled.
The cleaner edges provide smoother motion, kinda dreaming of 4X SR now tbh and SuperSampling too while we're at it http://forum-images.hardware.fr/images/perso/sniperlk.gif

New GPU's will be out soon and I'd happily use them for outrageous SS SR.
Ok, thanks. I can confirm your results. That's too bad, though, it means higher GPU consumption. And SuperRes is not exactly cheap, anyway... :(

I'm experiencing a strange v-sync issue with madVR.
This is totally outside of madVR's control. Most probably a GPU driver issue.

Madshi, will image enhancements be shown on the OSD? Or perhaps set an option to allow them to be shown?
Why? There are a million settings that I could show in the OSD. Why would the image enhancements be special to warrant being added to the OSD? I've added the scaling algos there because it's not always intuitive which decisions madVR makes there when using weird doubling settings etc. But for image enhancements there's nothing special or weird about what madVR is going.

Image thinning produces very similar results to awarpsharp but I still prefer the lines it makes (tends to bow things out a bit less) is it possible for the line thinning can be tweaked to produce perhaps similar results? Maybe even something the user could set?
Can you post a screenshot set which shows this quality difference? Please post the original unsharpened screenshot, then the one using awapsharp and the one sharpened by madVR. Try to tune the settings to get identical results. Which might not be easy. Obviously with higher strength there's a higher feature distortion.

Oh and if you got a chance to add line darkening to the toolbox that's be great.
I'm not sure about that. It sounds like a very specialist algorithm only useful to few people? All the other algos should be useful for everybody. "thin edges" might be an algo which was originally aimed at anime, but it's deadly useful for refining upscales, too, so it's really a very general purpose algo. While line darkening seems to be only useful for anime content to me? This might be an algo most suitable to be run in custom shaders?

[chros]

In my case, on my TV I find 2.2 too bright overall, and 2.4 (even though it comes out of black very quickly, shadow areas are far too bright with both 2.2 and 2.4) a bit too dark overall and a bit too contrasty. Setting the TV to 2.2 and having MadVR apply a 2.35 gamma curve seems just about perfect.
One thing popped in my mind: do you have a "black level" setting on your TV? It usually has the values Low/High. Isn't it set to High for you? Then use Low. And set levels to PC Levels in madvr for your display (or custom levels, as huhn suggested).

Btw: what levels should be set in this case in Lav video decoder. PC or Untuoched? Or does it matter?
I'm using PC now, I have tried untouched but I haven't seen a difference.

why do a double gamma correction that shouldn't even fix issue like that?

and don't even assume you get the gamma you setup in a TV. without calibration and a meter you don't know what you get and you can assume you have more of an S curve gamma than anything else.

for a dark room you should aim at 2.4 anyway.
Yep, that was a good question :) I haven't known about the purpose of the custom levels option.
So, I tried it out (instead of changing gamma level to 2.15 in madvr): I had to set black to 1, and all is good. Thank you! The picture is slightly darker now.

Since we are here let me ask you 2 more things. (I don't have meters, I did "calibrate" my display using the mp4 avshd mp4 basic and some other files (for the colors with the help of the built in primary color filters of the TV, that means I haven't changed the settings for grayscale calibration)).

1. what would you suggest to change for watching tv shows/movies during the day? Previously I always raised the backlight of the TV (from 40 to 80-90!), but now I thought maybe I should set the gamma on the TV to 2.2 (with the help of an another IFS user profile, so I can modify the necessary other settings as well (brightness, contrast, etc...)) and just switching between the 2 user profiles.

2. what would you set the color temperature to on the TV (without having done a proper calibration )?
Possible values: cool/medium/warm. The result of them exactly like you can see in the middle of this article. http://www.cnet.com/news/what-is-tv-color-temperature-and-why-does-it-matter/
I used medium for couple of months now, but yesterday night I tried out warm and it wasn't so bad as was it before (thanks to the modifications that you suggested).
I know that lot of people suggest to go for warm for tv series/movies (probably not for sport events), but it was strange for me all the time.
And last: can it be that you change the color temperature according to day or night watching? (I think no, but I'm not sure.)

Thanks for your help!

[leeperry]

I can confirm your results.
OK great, would there be room for further improvement there in your opinion? I guess another pass of SR on top wouldn't be possible without SuperSampling? http://forum-images.hardware.fr/images/perso/zytradance.gif

NNEDI3 is far more of a GPU hog than SR and I'm totally up for spending GPU load on SR coz again its effect on motion blur is astounding with Reclock, I currently have to zoom in native 1080p material one notch in order to trigger SR and 24p looks a hell lot smoother and focused :cool:

I can't bear watching non-SR footage anymore tbh, for instance motion on my HDTV tuner looks way too blurry to me now :angry:

Also eagerly awaiting having chroma SR updated but I guess that'll happen soon or later.

:thanks:

[derpycat]

Hi I'm having a fairly new issue, though not sure when it started. For some reason 29.796fps videos have 16.68ms intervals instead of the 33.36ms you'd expect, and so I'm getting frame drops even with an average rendering time of 16.6ms. Is this a new behaviour?

[Murmeltier]

Your wish is my command: http://screenshotcomparison.com/comparison/150162

All pictures as a .zip file (https://mega.nz/#!fhBlxaqY!fv2e7e_bTuMTG9QXNvCgAXlwjxWChkQ_8puMZT2R4Bg), using .13 with sxbr25+GL SR(St4/Sh2/R0.66), no other refinement was checked and I didn't try LL. Dithering was disabled.
The cleaner edges provide smoother motion, kinda dreaming of 4X SR now tbh and SuperSampling too while we're at it http://forum-images.hardware.fr/images/perso/sniperlk.gif

New GPU's will be out soon and I'd happily use them for outrageous SS SR.

Now a direct comparison with NNEDI128 (luma), NNEDI64 (chroma) image doubling and activated "enhance detail" (value = ~0.4) + "crispen edges" (value = 1.1) upscaling refinements in conjunction with possible Catmull-Rom (AR+LL) downscaling and disabled SuperRes would be interesting.



Ok, thanks. I can confirm your results. That's too bad, though, it means higher GPU consumption. And SuperRes is not exactly cheap, anyway... :(


To make it short. Isn't NNEDI128 image doubling without SuperRes still better than sxbr25 image doubling + an big portion of SuperRes sugar and almost equally hungry when it comes to performance?

[huhn]

is the file interlaced?

[derpycat]

is the file interlaced?

Ah, that was it. I had deinterlacing if in doubt, but the files weren't interlaced. Thanks

[aufkrawall]

To make it short. Isn't NNEDI128 image doubling without SuperRes still better than sxbr25 image doubling + an big portion of SuperRes sugar and almost equally hungry when it comes to performance?
There'd have to be a lot of SuperRes passes to get close to NNEDI3 128 performance needs.
I suppose it's not realistic and instead I'd limit SuperRes passes/strength (and NNEDI3 neurons to 64) and maybe apply a bit of "sharpen edges".
sxbr 25 is extremely soft, I think it doesn't make sense at all to use it if you have a GPU that's fast enough for at least 32 neurons of NNEDI3. I suspect there are quite some unwanted side effects when you try to sharpen sxbr 25 with "infinite" SR passes to a decent level.

Btw: I really hope "sharpen edges" won't be more consuming than Adaptive Sharpen in the end. :)

[leeperry]

Now a direct comparison with NNEDI128 (luma), NNEDI64 (chroma) image doubling and activated "enhance detail" (value = ~0.4) + "crispen edges" (value = 1.1) upscaling refinements in conjunction with possible Catmull-Rom (AR+LL) downscaling and disabled SuperRes would be interesting.

To make it short. Isn't NNEDI128 image doubling without SuperRes still better than sxbr25 image doubling + an big portion of SuperRes sugar and almost equally hungry when it comes to performance?
The thing is that SR is a lot smarter than NNEDI3 both FWIR and to my eyes: http://forum.doom9.org/showpost.php?p=1721618&postcount=90

I also don't like how NNEDI3 makes my HD7850 fans whine due to the high load and I especially don't like the artificial looking "thin lines" color NNEDI3 adds to all edges, I'll give you that sxbr+SR might look a little thick but the new "thin edges" with a low value might help.

TBH I only bother posting screenshots comparisons when madshi demands them as I trust my eyes more than anything else and again to my eyes nothing can replace SR so it's a matter of feeding it a scaler that's not über-sharp to begin with. I've posted a .zip file with everything you need if you feel like tinkering ;)

[ace960]

When next file in a directory is played then the audio is played only. The video freezes at the previous file.

Does the media player still react to mouse and keyboard events? E.g. can you press the stop button and it reacts?


The media player reacts. When I press stop button then it is pressed and the sound of the actual file stops playing. The video remains frozen until I close mpc-hc.

I would also add information that the system have two displays. One is the primary PC monitor. The second one is TV where the movies are played fullscreen. I operate mpc-hc from the main PC monitor taskbar (for this kind of test).

I also updated mpc-hc from 1.7.9.202 to 1.7.9.213. It did not change the behavior.

[mcn]

@mcn

try overlay mode may fix the issue.

Unfortunately that wasn't it.
Thanks though.


This is totally outside of madVR's control. Most probably a GPU driver issue.

I see, thanks.

[huhn]

so you get tearing with overlay mode?

[har3inger]

Unfortunately that wasn't it.
Thanks though.



I see, thanks.

Did you try playing with General Settings, checking/unchecking "use separate device"?

[markanini]

Crispen edges can look a little rough/aliased, which enhance detail brings out further. Nevertheless the two make my favorite combo with enhanced detail enabled under image enhancements. Thin edges seems to help with light haloing, looks better in upscaling refinement. In earlier versions I would normally prefer finesharp swhitching to lumasharpen if I saw artefacts. My sources are mostly live action. I always have Superres enabled in upscaling refinement.

[Warner306]

Crispen edges can look a little rough/aliased, which enhance detail brings out further. Nevertheless the two make my favorite combo with enhanced detail enabled under image enhancements. Thin edges seems to help with light haloing, looks better in upscaling refinement. In earlier versions I would normally prefer finesharp swhitching to lumasharpen if I saw artefacts. My sources are mostly live action. I always have Superres enabled in upscaling refinement.

I find Sharpen Edges + Enhance Detail is a superior combination than Crispen Edges + Enhance Detail. Edge enhancement is less obvious.

[aufkrawall]

I don't like crispen edges either, with it alone image does look hardly shaper to my eyes and I don't find combinations with it very useful (rather combine SuperRes with a strength of 1 or 2 with sharpen edges and line thinning). It's also the only one of the new sharpeners that increases brightness.
(while sharpen edges on the other hand can decrease it a tiny bit).

[ryrynz]

Why? There are a million settings that I could show in the OSD. Why would the image enhancements be special to warrant being added to the OSD?


It's useful at a glance.. Could you at least state the profiles that are running?


Can you post a screenshot set which shows this quality difference? Please post the original unsharpened screenshot, then the one using awapsharp and the one sharpened by madVR. Try to tune the settings to get identical results. Which might not be easy. Obviously with higher strength there's a higher feature distortion.


Yeah I know. I'm not sure I'd call it necessarily a quality difference though.. but I'll see what I can do.


I'm not sure about that. It sounds like a very specialist algorithm only useful to few people? All the other algos should be useful for everybody. "thin edges" might be an algo which was originally aimed at anime, but it's deadly useful for refining upscales, too, so it's really a very general purpose algo. While line darkening seems to be only useful for anime content to me? This might be an algo most suitable to be run in custom shaders?

Fair enough..

[mcn]

so you get tearing with overlay mode?

Yes, I occasionally get tearing on applications that don't use madVR, like Outlook.

Did you try playing with General Settings, checking/unchecking "use separate device"?

Thanks for the suggestion.

Yes, I tried all the four possible combinations involving use a separate device for presentation and use a separate device for DXVA processing.
Even these didn't make a difference.

[meetajhu]

madVR crashing MPC-HC x64 after Windows 10 fall update(version 1511). Is there any fix?

[truexfan81]

madVR crashing MPC-HC x64 after Windows 10 fall update(version 1511). Is there any fix?

are you running nvidia graphics? i ask cause on my windows8.1 everytime madvr crashes mpc-hc the traceback shows it to be an issue with the nvidia drivers

[Murmeltier]

I really can't spot an enormous difference while comparing "crispen edges" and "sharpen edges" with 1080p content on a 1080p display when it comes to sharpness. It looks like there's indeed a difference when it comes to brightness. "Crispen Edges" seems to brighten the image, "Sharpen Edges" does nothing of that kind and that's why I personally favor "Sharpen Edges". "Shapren Edges" produces clearly a more natural looking image to me. There's just an enormous drawback. The algorithm consums about 7ms of render time on my GTX 960 (not OC) and that's just too much in combination with my other settings I don't want to adjust.

It would be great if it is possible to optmize the algorithm so that is just needs about 3ms of render time, maybe 4ms. Hey, at least I'm free to dream.

I have a question reagarding DXVA2-copyback decoding. Would there be a huge performance gain if I replace - say a DD3-1300 CL 9-9-9-24 RAM - with DDR3-1600 7-8-8-24 modules?

[meetajhu]

are you running nvidia graphics? i ask cause on my windows8.1 everytime madvr crashes mpc-hc the traceback shows it to be an issue with the nvidia drivers

I found the problem. I have 144hz 3d vision monitor and after the new update 3d mode was ON under Windows 10 display settings(not Nvidia control panel) this was causing the player to crash.

[ryrynz]

I have a question reagarding DXVA2-copyback decoding. Would there be a huge performance gain if I replace - say a DD3-1300 CL 9-9-9-24 RAM - with DDR3-1600 7-8-8-24 modules?

I very much doubt it. Try downclocking or overclocking your RAM to get an idea of what you would stand to gain.

[huhn]

Yes, I occasionally get tearing on applications that don't use madVR, like Outlook.



Thanks for the suggestion.

Yes, I tried all the four possible combinations involving use a separate device for presentation and use a separate device for DXVA processing.
Even these didn't make a difference.

should be a GPU driver issue nothing you can do about.

[aufkrawall]

Some more thoughts on SuperRes radius:
I think one should simply use a lower SR strength for very high scaling factors, e.g. not above 2.
With it, I think aliasing is not an issue with real content in the wild and with high scaling factor, the effect of SR subjectively looks stronger to me in general.
So imho a radius of 0.66 is fine in general and if one wants less aliasing but still a nice sharpness boost, SR strength of 2 should be combined with the new sharpen options (at least once they are more affordable).
I would find it really bad if the radius was higher, as I really like the naturally sharp impression when doing 1080p -> WQHD. I suppose a radius of 0.66 also looks fine with 1080p -> 4k, as the aliasing seems to depend on a very low pixel count of thin structures.

[markanini]

I find Sharpen Edges + Enhance Detail is a superior combination than Crispen Edges + Enhance Detail. Edge enhancement is less obvious.

I don't like crispen edges either, with it alone image does look hardly shaper to my eyes and I don't find combinations with it very useful (rather combine SuperRes with a strength of 1 or 2 with sharpen edges and line thinning). It's also the only one of the new sharpeners that increases brightness.
(while sharpen edges on the other hand can decrease it a tiny bit).

If you like sharpen edges cause it's more well behaved then we there's no disagreement. Subjectively crispen edges is more visually pleasing, especially if the source is clean.

[mogli]

That's a cosmetical OSD bug. Will be fixed in the next build. You can safely ignore this, it has no meaning.
I'm not so sure about that, but let's wait and see.

There's another isssue with those interlaced DVDs since the introduction of the automatic switch to DXVA scaling:
When playback starts the clock deviation is heavily off, maybe 10^6 % or so, and the DVD essentially hangs for 10 to 20 sec until this value goes down to an acceptable level and plays fluidly thereafter.

[Sp00kyFox]

hi there madshi. first time I'm posting here but I'm using your VR for a long time now. I was playing around with the new implemented sharpen agorithms and so far I'm pretty satisfied with sharpen and crispen, both at default values. the effect with superres is too subtle though for my taste, but maybe my display resolution is too low for the filter to be useful. I usually watch 720p content with a native screen resolution of 1680x1050. regarding luma and adaptive sharpen, I also don't see much use for them. I can't imagine content where I would prefer them over the new four ones.

anyways, I wanted to draw some attention to a new developed downscaling algorithm by ETH Zurich which is called "Perceptually Downscaling" which gets pretty amazing results:
https://graphics.ethz.ch/~cengizo/imageDownscaling.htm

I think that could be a gread addition to madvr. according to the paper the performance of is pretty good (they implemented it with matlab) and it should be possible to execute it in realtime. it seems that they didn't publish the code though but the algorithm is explained in the paper and presented in pseudo code, it only consists of convolutions and sums.