madVR - high quality video renderer (GPU assisted)

[Sebastiii]

Nice
Thanks.

[Frexxia]

Is there some sort of trick to get madTestPatternSource working? I've installed it, and 'The DirectShow filter "madTestPatternSource.ax" was registered successfully.' pops up, but it doesn't show up under External Filters in MPC-HC. If I try to open any of the *.ytp files I just get "Cannot render the file".

I'm sure there is something obvious I'm missing.

[thewildsun]

Originally Posted by austinminton

I had the same problem and I fixed it by changing my avr.

Such problems are quite specific to the setup and I had to go through a lot of trial and error before it got fixed.

Thanks for the suggestion. I tried a new AVR (from Denon to Marantz), and the problem actually became worse -- there was a hum even when the HTPC was idle. So it turned out this was a ground loop issue afterall, and perhaps the Marantz is more sensitive to the issue. I tested a cheater plug on the HTPC, and ran it, the AVR, and the amp all on separate power strips, and the hum went away. Going back to the Denon instead of the Marantz with this change, the GPU still whines but it doesn't get to the speakers anymore. I still need to figure out a permanent solution that doesn't involve cheater plugs, though.

Originally Posted by madshi

Might make sense to double check that you don't have DPC latency issues. See here:

http://www.thesycon.de/eng/latency_check.shtml

I had these issues (plus occasional blue screens) on my rather old HTPC when I allowed the HTPC to dynamically clock the CPU down. Fixing the CPU at normal clocks all the time solved the issue for me.

Thanks for the suggestion as well. I did try checking DPC latency and there didn't seem to be any issues. As you can see above, this problem was most likely caused by a ground loop issue. The GPU itself does still whine, but perhaps that is normal GPU coil whine, which I can live with as long as it doesn't travel to the speakers anymore.

Also, I just wanted to report that it turns out that even with the updated PSU, I started getting the frame drop issue I mentioned before once again, but I was finally able to determine that it was caused by my anti-virus, BitDefender 2018. I have no idea why it does this, or why it causes frame drops only after playing something perfectly smooth for several minutes, but I was able to fix it by whitelisting the directories for PotPlayer, madVR, and LAV filters. Hope this helps someone.

[huhn]

Originally Posted by madshi

IMHO the pixart image creator's intent was not to create low-res highly aliased images. They simply had no other choice, because the game consoles at the time didn't support any higher resolution. So they did the best they could, setting every pixel carefully by hand. However, if they had had the chance to use a higher resolution, they probably would have, and in the process they would have created images that would have been much less aliased.

Now it's a valid question if NGU AA is able to "interpret" the images in such a way that brings them nearer to the original image creator's intent or farther away. This is a topic that would probably be valid grounds for heated discussions. Personally, when re-playing old pixeart adventure games, I'd much rather have NGU AA upscaled images than nearest neighbor. But I suppose every user has a right to decide for himself. Generally saying that upscaling pixart with NGU AA does not honor the image creator's intent is IMHO not fair, at least I don't agree with that.

while it may not be intended back in the days pixel art games are created even today. and i'm pretty sure it is intended now.

Who in this thread cares about "most scalers"? The key question is what madVR can do with it.

FWIW, DVD sources are not even encoded in their native aspect ratio, so it's impossible to display DVDs without scaling them in at least one direction!

yes we have to scale them now but that was not the case with old CRT screen or even to some degree with LCDs and not the important part here it is the huge scaling factor with UHD screens.
it's not easy to make UHD picture for comparison they are usually to huge for image hoster.

here some screens:
native: https://abload.de/img/title06.mkv_snapshot_1eue5.png
NGU sharp high: https://abload.de/img/ngusharp71uf0.png
NGU AA: https://abload.de/img/nguaajau63.png
lanczos3: https://abload.de/img/lanczosnuuja.png

NGU sharp does some wonders on the background tiny blue red flags and half side of the white helmet(and way more) but it highlights compression artefacts and it is quite odd that some image parts are not sharp at all and some look pretty ok.
NGU sharp is known to work best with good source this source has nothing todo with good.

every other used scaler is significantly less sharp but has a more balanced image.
is NGU sharp sharper than the source image maybe? if not every other scaler makes it less sharp than the source.
scaling an image like this is at best a trade.

i used NGU high, let madVR decide except for downscaling i used SSIM 1D100 AR.

madVR 0.92.7 RCA 2 is effect on this image but it is not used because this is about scaling.

[xabregas]

Originally Posted by madshi

Well, I've been recommending for months (years?) now to use Windows 8.1 instead of Windows 10, but only very few people listen to me.

That said, with 385.xx drivers it might have worked ok.

People want to keep up to date and sometimes things get worst.

Audio speaking. Since XP audio only got worst on new windows os. And windows 8.1 probably has the best desktop composition of them all.

I think windows 10 was made for centennials. Z people love to see their desktops full of flying windows or notes or photos. Videos coming from right to the left. I cant stand it. I like to control EVERYTHING on my machine. With windows 10 thats IMPOSSIBLE.

Anyways, madshi keep up the good work. We all love you. Its something to praise what you do for free.

[pirlouy]

Originally Posted by madshi

Well, I've been recommending for months (years?) now to use Windows 8.1 instead of Windows 10, but only very few people listen to me.

'cause you forgot to explain in detail what is that bad with Windows 10 (at least in 1st page).
I'm quite confident your reason is justified, but as a basic user, I would need a really good reason not to use last Windows version.
And it's certainly not xabregas's nonsenses which will make me change my mind. For now, I have 0 problem with MPC+madVR or mpv.

[leeperry]

RCA@1 in combination with NGU Sharp "High" looks like a total winner on 960*540@1080p

[Anima123]

Madshi,

Is RCA applied before IVTC, right? I wonder if you can let RCA be applied AFTER IVTC, since IVTC is really useful when playing back videos with 59.976 or 60 fps, in which case the original video can be decimated, to save limited GPU power to more crucial tasks.

Best Regards,
Anima123

[huhn]

IVTC is done with the CPU so you can blindly assume it is done before.

using RCA on it before IVTC should make it pretty hard to play IVTC.

[ryrynz]

Originally Posted by Frexxia

Is there some sort of trick to get madTestPatternSource working? I just get "Cannot render the file".

Test with a 32 bit media player.

[jkauff]

I bought a very nice expanded color range 4K monitor, which arrived Saturday. It's going back Monday because my GTX 1060 6GB couldn't handle Blu-ray doubling without major frame drops.

I'll try again next time I upgrade the video card. Madshi was right, the picture was gorgeous.

EDIT: Just noticed the GTX 1080 11GB has come down in price. I assume that would do the job, but does anyone here know for sure?

[huhn]

my 1060 can do ngu sharp very high with normal chroma when upscaling 1080p23 to UHD.

so are you sure you didn't forget to change the nvidia power setting to adaptive?
and if you want tell people want you want to do no one can tell you if a 1080 can do X.

[jkauff]

Originally Posted by huhn

my 1060 can do ngu sharp very high with normal chroma when upscaling 1080p23 to UHD.

so are you sure you didn't forget to change the nvidia power setting to adaptive?
and if you want tell people want you want to do no one can tell you if a 1080 can do X.

I guess you have a better 1060 than I do. Mine won't do doubling at the lowest settings without dropping frames.

All I want to do is simple doubling with NGU. And yes, I'm on Adaptive power setting.

[Razoola]

Originally Posted by jkauff

I guess you have a better 1060 than I do. Mine won't do doubling at the lowest settings without dropping frames.

All I want to do is simple doubling with NGU. And yes, I'm on Adaptive power setting.

I also feel there might be something off in your settings. I have a GTX980 and that can do doubling with NGU sharp high no problem. These cards are about even in preformance.

[huhn]

NGU low is pretty much free so can you make a screen of the OSD?

[Frexxia]

Originally Posted by ryrynz

Test with a 32 bit media player.

Thanks!

There should be a warning that it doesn't work with 64 bit players somewhere.

[jkauff]

Originally Posted by huhn

NGU low is pretty much free so can you make a screen of the OSD?

I found the problem. I had accidentally left RCA and RRN enabled (at 1). Disabling them got rid of the drops, so all is well now.

[ashlar42]

Originally Posted by madshi

Not true at all. Actually, the better the source quality, the better NGU Sharp looks. E.g. have you seen these screenshots yet?

http://www.avsforum.com/forum/24-dig...l#post54983374

These are from one of the best UHD Blu-Rays (in terms of detail and resolution) we have so far, and madVR's 1080p -> UHD upscale gets remarkably close to the UHD image, when talking about edges. Of course madVR can't invent texture detail which isn't there in the 1080p image. But edges upscaled to 4K look *so* much better than when displaying the 1080p image untouched.

I surrender! I accept your opinion as true. And I'm not ironic or anything. I respect your knowledge in this field and rest assured that whenever I'll get the money to upgrade to 4K madVR will be there with me in getting the best quality possible on screen.
(I'll go have a look on AVS, thanks for the link)

[madshi]

Originally Posted by huhn

while it may not be intended back in the days pixel art games are created even today. and i'm pretty sure it is intended now.

Maybe, I don't know. Are these pixel art games created today as aliased as those old low-res pixel art games? Or do they have smooth "anti-aliased" edges now?

Originally Posted by huhn

yes we have to scale them now but that was not the case with old CRT screen or even to some degree with LCDs and not the important part here it is the huge scaling factor with UHD screens.
it's not easy to make UHD picture for comparison they are usually to huge for image hoster.

here some screens:
native: https://abload.de/img/title06.mkv_snapshot_1eue5.png
NGU sharp high: https://abload.de/img/ngusharp71uf0.png
NGU AA: https://abload.de/img/nguaajau63.png
lanczos3: https://abload.de/img/lanczosnuuja.png

NGU sharp does some wonders on the background tiny blue red flags and half side of the white helmet(and way more) but it highlights compression artefacts and it is quite odd that some image parts are not sharp at all and some look pretty ok.
NGU sharp is known to work best with good source this source has nothing todo with good.

every other used scaler is significantly less sharp but has a more balanced image.
is NGU sharp sharper than the source image maybe? if not every other scaler makes it less sharp than the source.
scaling an image like this is at best a trade.

i used NGU high, let madVR decide except for downscaling i used SSIM 1D100 AR.

madVR 0.92.7 RCA 2 is effect on this image but it is not used because this is about scaling.

Well, first of all we were originally talking about 1080p -> 4K upscaling, which is not the same huge upscaling factor as DVD -> 4K, but anyway.

I've one problem with your post: It's a comparison of NGU AA vs NGU Sharp vs NGU Lanczos. But it's *not* a comparison of NGU vs unscaled. Some users might think it is, because you provided the original untouched source image, but it's not. Why? Because in order to compare 2 images, both images need to have the same *viewing angle*. If you compare 2 images where one image covers a much smaller angle of your view than the other, then the one with the smaller viewing angle will almost always look sharper and less artifacty. But it's not a fair comparison, because we're not discussing watching 1080p content on a 25" 1080p display, compared to watching 1080p content on a 50" 4K display! What we're discussing is watching 1080p content on a 50" 1080p display, compared to watching 1080 content on a 50" 4K display. Do you agree? The 1080p and 4K displays will have the same viewing angle. So in order to do a fair comparison we also need to compare "screenshots" with the same viewing angle.

So how can we do a fair comparison? It's actually simple: We use a digicam/DSLR to shoot an image of the unscaled source, and we shoot it in such a way that the unscaled source covers the whole digicam frame. Then we shoot an image of the upscaled image, again in such a way that the upscaled image covers the whole frame. This way we get a relatively fair comparison how the upscaled vs unscaled images look like, when seen at the same viewing angle.

Unfortunately there's a catch here: Your unscaled DVD image has a distorted aspect ratio, so I had no choice but to correct the aspect ratio by upscaling. I've used the same algos for AR correction, and for upscaling.

Here are the results, straight out of the digicam, no processing other than simple cropping:

digiCamNativeRes.png - | - digiCamUpscaled.png

Please make sure your browser doesn't zoom these!

Due to technical limitations (physical resolution of my monitor) it's a relatively small upscaling factor, but I'm pretty sure the results would have been identical (or even more in favor of the upscaled version) with a bigger upscaling factor.

Please feel free to repeat this test yourself!

Originally Posted by Anima123

Is RCA applied before IVTC, right?

Originally Posted by huhn

IVTC is done with the CPU so you can blindly assume it is done before.

^ IVTC first, RCA after.

Originally Posted by ashlar42

I surrender! I accept your opinion as true. And I'm not ironic or anything. I respect your knowledge in this field and rest assured that whenever I'll get the money to upgrade to 4K madVR will be there with me in getting the best quality possible on screen.

Thanks, appreciate that. Also check out my reply to huhn above, it might confirm things for you.

FWIW, there is more to this than just the question whether upscaling is useful or not. E.g. if the 1080p display has better native contrast, overall you might still be better off with the 1080p display than with the 4K display. So 4K isn't always better than 1080p.

[huhn]

Originally Posted by madshi

Maybe, I don't know. Are these pixel art games created today as aliased as those old low-res pixel art games? Or do they have smooth "anti-aliased" edges now?

this game got sales in the millions:
https://abload.de/img/stardewm6u3m.png

Well, first of all we were originally talking about 1080p -> 4K upscaling, which is not the same huge upscaling factor as DVD -> 4K, but anyway.

I've one problem with your post: It's a comparison of NGU AA vs NGU Sharp vs NGU Lanczos. But it's *not* a comparison of NGU vs unscaled. Some users might think it is, because you provided the original untouched source image, but it's not. Why? Because in order to compare 2 images, both images need to have the same *viewing angle*. If you compare 2 images where one image covers a much smaller angle of your view than the other, then the one with the smaller viewing angle will almost always look sharper and less artifacty. But it's not a fair comparison, because we're not discussing watching 1080p content on a 25" 1080p display, compared to watching 1080p content on a 50" 4K display! What we're discussing is watching 1080p content on a 50" 1080p display, compared to watching 1080 content on a 50" 4K display. Do you agree? The 1080p and 4K displays will have the same viewing angle. So in order to do a fair comparison we also need to compare "screenshots" with the same viewing angle.

So how can we do a fair comparison? It's actually simple: We use a digicam/DSLR to shoot an image of the unscaled source, and we shoot it in such a way that the unscaled source covers the whole digicam frame. Then we shoot an image of the upscaled image, again in such a way that the upscaled image covers the whole frame. This way we get a relatively fair comparison how the upscaled vs unscaled images look like, when seen at the same viewing angle.

Unfortunately there's a catch here: Your unscaled DVD image has a distorted aspect ratio, so I had no choice but to correct the aspect ratio by upscaling. I've used the same algos for AR correction, and for upscaling.

Here are the results, straight out of the digicam, no processing other than simple cropping:

digiCamNativeRes.png - | - digiCamUpscaled.png

Please make sure your browser doesn't zoom these!

Due to technical limitations (physical resolution of my monitor) it's a relatively small upscaling factor, but I'm pretty sure the results would have been identical (or even more in favor of the upscaled version) with a bigger upscaling factor.

Please feel free to repeat this test yourself!

my real problem is scaling in general is not a pure positive effect which will make images better.

the image i post you looks odd with NGU sharp because some parts get sharp after scaling and other don't and this results in a odd look just look at the ship the "mast" they are super sharp compared to the rest of the image which is odd because it is clearly not in focus.