madVR - high quality video renderer (GPU assisted)

[Razoola]

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?

The purists of pixel art games would not use any type of smoothing, they would much rather have the jaggies. The only thing that would interest them would be scanline filters maybe providing it does not compromise pixel accuracy.

I'm not sure any type of generated scanlines (like done in emulators) would be wanted in madVR unless there is video content out there that would look better on an old Low Khz arcade game type monitor than todays panels.

[madshi]

Originally Posted by huhn

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

Euuw, not sure why anybody would create such aliased ugliness today.

Originally Posted by huhn

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.

You are correct in saying that upscaling algorithms aren't perfect. But I'm sorry to say, you're completely missing the point. You are comparing images with a different viewing angle!!! Because you're looking at the upscaled images in 100% view on your PC, and comparing them to the original unscaled image in 100% view on your PC. Until you understand this crucial point there's no reason to discuss any further. Please do a fair apples to apples comparison, using the same viewing angle for unscaled vs upscaled images, and I'm sure you'll finally see that upscaled is overall noticeably better, even though upscaling algorithms are admittedly not perfect.

[Razoola]

Originally Posted by madshi

Euuw, not sure why anybody would create such aliased ugliness today.

There is big money to be made here, esp games on old arcade systems. Games by NG DEV.TEAM for Neo Geo as an example sell for over 400 euro a pop. Those are at a resolution of 320x224 and look graphically really impressive...

https://www.youtube.com/watch?v=TMix_9cesik

[mparade]

Originally Posted by madshi

It seems to be a case of 1080p movie on 1080p screen, so why should madVR double? Did you activate supersampling? If so, it might make sense if you make screenshots of your doubling settings page. Also please make sure that the profile you want to be active is actually active (shown in "bold" font) in this situation.

madVR should double because I have HD material this case, my display is with 4K resolution and I do not want the display to upresize the source. I haven't activated supersampling, I would like madVR to double the 1920x1080p source to 4K resolution. The correct profile is actually active during play. Screenshots were being uploaded here:

http://s1.toldacuccot.hu/letoltes?si...fb453009761334

[mparade]

Originally Posted by madshi

Can't say anything about that without seeing a screenshot of what you mean. madVR is usually not responsible for the player GUI, unless the player uses the madVR GUI/OSD APIs.

I am using JRiver 32bit with it's own madVR and LAV version installed. Screenshot what I meant is uploaded here:

http://s1.toldacuccot.hu/letoltes?si...6b50e77ac24d11

GUI is mixed up only after returning from HD play. Returning from 4K play is without any GUI problem, as it can be seen from the screenshots above.

[mparade]

Originally Posted by madshi

Which issue are you talking about? Not doubling? Or mixed up GUI?

mixed up GUI. After I had changed the renderer in JRiver from madVR (Red October HQ) to the standard one (Red October Standard), there was no mixing up anymore.

[Werewolfy]

Originally Posted by madshi

madVR v0.92.8 released

http://madshi.net/madVR.zip

* added 3 new "reduce compression artifacts" strength steps between 1-4
* removed deblock option "don't reduce texture detail", wasn't useful
* replaced old "reduce random noise" strengths 1+2 with a new strength 1
* fixed: key for increasing "reduce random noise" strength stopped at 5

Not much time for madVR atm, so just some smaller algo tweaks.

Thanks for the new "reduce compression artifacts" strength steps. I can confirm right away that even strength 1 does make a difference without damaging the picture.

Concerning "reduce random noise", strength 1 is a good choice now if you want to remove a little bit the noise without the details destruction of higher strengths. But it's still not my cup of tea, too much details are blurred during the process.

RRN off vs RRN 1

Look at the writing on the sheet of paper. I suppose there's nothing you can do with this type of algo. I hope that someday you will bring multi-frame noise filter, that would be awesome

Originally Posted by madshi

Euuw, not sure why anybody would create such aliased ugliness today.

Well, it's far from being the only example. There are still A LOT of "such aliased ugliness" today.

Terraria
TowerFall Ascension
Hyper Light Drifter
Shovel Knight

I could show you a lot more videogames like that, that's just some examples. Their aliased look are clearly made on purpose and any upscalings other than Nearest-Neighbor alters the developper's intention. I'm a big fan of NGU for movies and animes but I will never use any "real" upscaling on pixel arts. I hate what they do on this type of content.

I hope that someday you will bring back Nearest-Neighbor into the GUI, it's the way these games are meant to be played and obviously watched.

[huhn]

Originally Posted by madshi

Euuw, not sure why anybody would create such aliased ugliness today.

it made him a couple mills so not sure i would argue with that.

You are correct in saying that upscaling algorithms aren't perfect. But I'm sorry to say, you're completely missing the point. You are comparing images with a different viewing angle!!! Because you're looking at the upscaled images in 100% view on your PC, and comparing them to the original unscaled image in 100% view on your PC. Until you understand this crucial point there's no reason to discuss any further. Please do a fair apples to apples comparison, using the same viewing angle for unscaled vs upscaled images, and I'm sure you'll finally see that upscaled is overall noticeably better, even though upscaling algorithms are admittedly not perfect.

sorry your digicam example doesn't count too your screen is scaling so we can't do this properly except with a CRT and a lot of work.

so let's compare lanczos with NGU sharp. while NGU sharp is i guess generally better than lanczos is looks like it is getting out of focus because soem part get really sharp and other not. so i'm just choicing an lesser evil that's not a general increase in image quality in my book. do the good out weight the bads i'm pretty sure it is still just a trade in my book.

i was just disagreeing that upscaling is generally better.
i have to upscale it anyway so...

[madshi]

Originally Posted by Werewolfy

Concerning "reduce random noise", strength 1 is a good choice now if you want to remove a little bit the noise without the details destruction of higher strengths. But it's still not my cup of tea, too much details are blurred during the process.

Yeah, it is what it is.

Originally Posted by Razoola

There is big money to be made here, esp games on old arcade systems. Games by NG DEV.TEAM for Neo Geo as an example sell for over 400 euro a pop. Those are at a resolution of 320x224 and look graphically really impressive...

Originally Posted by Werewolfy

Well, it's far from being the only example. There are still A LOT of "such aliased ugliness" today.

Terraria
TowerFall Ascension
Hyper Light Drifter
Shovel Knight

I could show you a lot more videogames like that, that's just some examples. Their aliased look are clearly made on purpose

Originally Posted by huhn

it made him a couple mills so not sure i would argue with that.

Alright, you guys win. I still think that I would personally prefer having these upscaled with NGU AA instead of nearest neighbor, but I suppose it's a matter of taste.

Well, you probably won't want to constantly use nearest neighbor upscaling, so using a keyboard shortcut to activate it on demand should do the trick, shouldn't it?

Originally Posted by mparade

madVR should double because I have HD material this case, my display is with 4K resolution and I do not want the display to upresize the source. I haven't activated supersampling, I would like madVR to double the 1920x1080p source to 4K resolution.

Fair enough, but it seems that madVR thinks that the display mode is actually 1920x1080, and not 4K. I'm not sure if that's actually true, or maybe it's because of your very high DPI setting. Do you use refresh rate / display mode changing in either madVR or J.River somewhere? If so, try disabling that, then activate 4K mode manually, that might solve the problem.

Originally Posted by mparade

I am using JRiver 32bit with it's own madVR and LAV version installed. Screenshot what I meant is uploaded here:

http://s1.toldacuccot.hu/letoltes?si...6b50e77ac24d11

GUI is mixed up only after returning from HD play. Returning from 4K play is without any GUI problem, as it can be seen from the screenshots above.

It looks like the display mode switched from 4K to 1080p.

Originally Posted by huhn

sorry your digicam example doesn't count too your screen is scaling

No no no - my screen was *not* doing any scaling!!!

Maybe I didn't explain well. In order to shoot the "unscaled" image, I displayed your DVD image in MPC-HC with 100% view (called "normal size" in MPC-HC, resulting in 1024x576), using a display mode which matches the physical resolution of my LCD monitor. So every DVD pixel was displayed by one LCD pixel. Well, not really, because of Aspect Ratio correction. But if we ignore AR correction, it was a 1:1 pixel match. Then I took my digicam, used optical (!) zoom to shoot the image. So (other than AR correction) no digital zoom/scaling was applied by anyone. Not by madVR, not by the display, not by the digicam, not by the photo editor, by no one.

In order to shoot the "upscaled" image I simply switched MPC-HC from "normal size" to "touch from inside", changed the optical zoom in my digicam and shot the image again.

This way we have 2 images which very fairly compare unscaled vs upscaled, using the optical zoom in my digicam to get the same viewing angle. Got it?

Originally Posted by huhn

we can't do this properly except with a CRT and a lot of work.

Apart from AR correction, we absolutely can do it properly. See above.

Originally Posted by huhn

so let's compare lanczos with NGU sharp. while NGU sharp is i guess generally better than lanczos is looks like it is getting out of focus because soem part get really sharp and other not. so i'm just choicing an lesser evil that's not a general increase in image quality in my book. do the good out weight the bads i'm pretty sure it is still just a trade in my book.

i was just disagreeing that upscaling is generally better.

You cannot draw any conclusions without doing a proper apples to apples comparison, using the same viewing angle. I believe if you did a proper apples to apples comparison, you would completely change your mind.

[raymondjpg]

After applying the Windows 10 Fall Creators Update 1709, madVR display rate switching appears to be working with MPC-HC 1.7.13 64 bit, but stats appear to indicate an incorrect refresh rate. For example 23.976 fps source material is showing a refresh rate of 29.97. I have the same issue with both madVR v91.11 and v92.8.

Can someone point me to what is causing this, and suggest a solution? I haven't seen anything obviously wrong in the madVR setup. Display rate switching is set for 1920x1080 resolution.

TIA.

Edit: The issue appears to be a result of Intel HD Graphics Iris 640 driver v4749 installed by Microsoft's implementation of the Windows 10 Fall Creators Update 1709. While at some point this driver version may have been available for download from Intel I could find no trace of it there today. Reverting to an earlier driver version that I knew was working well with Windows 10 v1703 has resolved the stats issue. N.B. from what I could establish from the Intel HD Graphics control panel the madVR refresh rate switching was working correctly with the Iris 640 driver v4749, and it was the stats display that was not.

[ryrynz]

Originally Posted by Werewolfy

Concerning "reduce random noise" it's still not my cup of tea, too much details are blurred during the process.

Reduce Random Noise is more like, make this picture ever so slightly softer to hide what we think might be noise, need multi frame processing for the real deal so I won't personally ever use RRN.
Gimme a Spatial-Temporal Denoiser though and I'm all over it because those results are godly.

I did a fresh windows install but i still have some problems.
Using the option "Use D3D11..." solves the black screen, but when i leave fullscreen the player just shows the last image before it went fullscreen, but the audio is still playing.
Also when i hover the mouse on the seek bar the video starts dropping frames. Is there a solution or any setting to solve this?

EDIT: Did some testing here.
When i put mouse pointer at seek bar on fullscreen the present queue goes from 6-7/8 to 1-2/2, there is no dropped frames on stats but the video is clearly stuttering.
Also when i disable switch to matching display mode, the video plays normally on fullscreen or windowed, but if i enable the switch to match display mode, when i leave fullscreen at 24hz it just shows a static image on video player.

[khanmein]

Originally Posted by leandronb

I did a fresh windows install but i still have some problems.
Using the option "Use D3D11..." solves the black screen, but when i leave fullscreen the player just shows the last image before it went fullscreen, but the audio is still playing.
Also when i hover the mouse on the seek bar the video starts dropping frames. Is there a solution or any setting to solve this?

EDIT: Did some testing here.
When i put mouse pointer at seek bar on fullscreen the present queue goes from 6-7/8 to 1-2/2, there is no dropped frames on stats but the video is clearly stuttering.
Also when i disable switch to matching display mode, the video plays normally on fullscreen or windowed, but if i enable the switch to match display mode, when i leave fullscreen at 24hz it just shows a static image on video player.

https://blogs.windows.com/windowsexp...uild-17025-pc/

We fixed the issue where if you RDP into a PC running this build with certain GPU configurations, when you go to sign in to the PC locally it will appear stuck at a black screen with only the cursor available

We fixed an issue where toggling some DX9/DX10/DX11 games between windowed and fullscreen (for example using Alt + Tab) could result in the game window become black on certain PCs.

[huhn]

Originally Posted by madshi

No no no - my screen was *not* doing any scaling!!!

Maybe I didn't explain well. In order to shoot the "unscaled" image, I displayed your DVD image in MPC-HC with 100% view (called "normal size" in MPC-HC, resulting in 1024x576), using a display mode which matches the physical resolution of my LCD monitor. So every DVD pixel was displayed by one LCD pixel. Well, not really, because of Aspect Ratio correction. But if we ignore AR correction, it was a 1:1 pixel match. Then I took my digicam, used optical (!) zoom to shoot the image. So (other than AR correction) no digital zoom/scaling was applied by anyone. Not by madVR, not by the display, not by the digicam, not by the photo editor, by no one.

In order to shoot the "upscaled" image I simply switched MPC-HC from "normal size" to "touch from inside", changed the optical zoom in my digicam and shot the image again.

This way we have 2 images which very fairly compare unscaled vs upscaled, using the optical zoom in my digicam to get the same viewing angle. Got it?

the image is still upscaled by 42 % but i got what you did there.
and don't worry i watched them

just taking the image without AR correction and 2x NN and x2 NGU sharp should be better than this i guess.

even with NGU sharp i can see the gaps between the pixels.

You cannot draw any conclusions without doing a proper apples to apples comparison, using the same viewing angle. I believe if you did a proper apples to apples comparison, you would completely change your mind.

that was my conclusion.

so i'm just choicing an lesser evil that's not a general increase in image quality in my book. do the good out weight the bads i'm pretty sure it is still just a trade in my book.

and just to be sure you used add grain for NGU didn't you? the image looks a lot more details and that's something a scaler can't do.
there is a reason i ask you if you could add add grain for other scaler. not the creators intend but what ever doesn't look unnatural on an old DVD.
EDIT: if add grain was really used plz don't redo the screens that looks time consuming.

[ashlar42]

My two cents about nearest neighbor. If it's still in madVR and accessile through a programmable key, why not getting it back in order to use it for profiles? Maybe with a (useful for pixelart and old videogames) note beside it?

It's still there for image downscaling (where I don't see how and when it could make sense).

I couldn't care less, personally, about it. But if I watched, for whatever reason, lots of stuff like that, not having it available for profiles would irritate me.

[leeperry]

Originally Posted by ashlar42

not having it available for profiles would irritate me

open a paypal dispute and demand a full refund

[huhn]

the image upscale page is pretty much full and AFAIK this stuff is time consuming.

before something like this is added i would prefer a scaler like RAVU or something like that. pretty sure there are more user that find this scaler useful than NN.

[madshi]

Originally Posted by leandronb

I did a fresh windows install but i still have some problems.
Using the option "Use D3D11..." solves the black screen, but when i leave fullscreen the player just shows the last image before it went fullscreen, but the audio is still playing.
Also when i hover the mouse on the seek bar the video starts dropping frames. Is there a solution or any setting to solve this?

See khanmein's reply.

Also, which GPU and driver version is this? If Nvidia, DON'T use 387.xx or 388.xx, use 385.xx, please.

Originally Posted by huhn

the image is still upscaled by 42 % but i got what you did there.

Only in X direction. Y res stays the same. So it's much less than 42%. If you say an image was upscaled by 42%, usually people mean both X and Y.

Originally Posted by huhn

just taking the image without AR correction and 2x NN and x2 NGU sharp should be better than this i guess.

I agree that without AR correction would have been better. I'm pretty sure it would have increased the quality gap even further in favor of the upscaled image.

Why would 2x NN have been better, though? This whole discussion is about whether upscaling helps or harms image quality. So if I want to create digicam shots to prove my point, I *have* to create shots that show unscaled vs upscaled. If I had upscaled both (with NN vs NGU) I would have only compared NN vs NGU, I would not have compared unscaled to upscaled!

If you want to compare NN vs NGU, we can do that very easily and directly in madVR without using a digicam.

Originally Posted by huhn

even with NGU sharp i can see the gaps between the pixels.

Yes. Which is a good thing because it proves that my digicam isn't half bad. The photos faithfully show what my LCD monitor does, when displaying your DVD image unscaled vs upscaled. The photos show that displaying the content unscaled suffers more from the (sub)pixel grid of my LCD monitor than when using upscaling. The same effect exists when comparing 1080p content on a 1080p vs 4K display.

It's an important argument that displaying 1080p content untouched on a 1080p display doesn't faithfully represent the 1080p content, because many displays don't have perfectly rectangular pixels with 100% fillrate, and even more importantly, many displays have RGB subpixels that are positioned next to each other, so they don't cover the same area. This subpixel positioning can introduce additional aliasing/combing artifacts that aren't even in the source image, as seen in my digicam shots! So when showing a 1080p frame on a 1080p display, actually image quality gets lost due to the gaps between the pixels and the placement of the RGB subpixels. These problems of imperfect display technology are nicely reduced by upscaling the 1080p image to a higher resolution. And my digicam photos show that quite nicely.

Originally Posted by huhn

so i'm just choicing an lesser evil that's not a general increase in image quality in my book. do the good out weight the bads i'm pretty sure it is still just a trade in my book.

Yes, it's a trade. But the positives *FAR* outweigh the negatives, IMHO.

Originally Posted by huhn

and just to be sure you used add grain for NGU didn't you? the image looks a lot more details and that's something a scaler can't do.

Yes, but I also used it for the AR corrected image (which was AR corrected by doing NGU Sharp + Grain + SSIM1D downscaling), so it's still a fair comparison.

Originally Posted by huhn

there is a reason i ask you if you could add add grain for other scaler.

Yes, it's on my to do list.

[xabregas]

Originally Posted by leandronb

I did a fresh windows install but i still have some problems.
Using the option "Use D3D11..." solves the black screen, but when i leave fullscreen the player just shows the last image before it went fullscreen, but the audio is still playing.
Also when i hover the mouse on the seek bar the video starts dropping frames. Is there a solution or any setting to solve this?

EDIT: Did some testing here.
When i put mouse pointer at seek bar on fullscreen the present queue goes from 6-7/8 to 1-2/2, there is no dropped frames on stats but the video is clearly stuttering.
Also when i disable switch to matching display mode, the video plays normally on fullscreen or windowed, but if i enable the switch to match display mode, when i leave fullscreen at 24hz it just shows a static image on video player.

Same problem here. I reverted to windows 8.1. Guess what?? Problem solved and it didnt hurt. Windows 10 is a bug and when we think the bug is solved like in 1703 version they invent a fall creator evil update to show us the bugs they came up in 6 months of work and introduce new amazing options like the sleeping light and the gpu tab in task manager. In 4 months i believe everything will be fine and you will have 1/2 months of no bugs until the spring/summer creators update.

[huhn]

Originally Posted by madshi

Only in X direction. Y res stays the same. So it's much less than 42%. If you say an image was upscaled by 42%, usually people mean both X and Y.

a problem with % numbers . based on the total pixel count it is more than 42 % more pixels. sorry no clue how to make it more clear.

I agree that without AR correction would have been better. I'm pretty sure it would have increased the quality gap even further in favor of the upscaled image.

Why would 2x NN have been better, though? This whole discussion is about whether upscaling helps or harms image quality. So if I want to create digicam shots to prove my point, I *have* to create shots that show unscaled vs upscaled. If I had upscaled both (with NN vs NGU) I would have only compared NN vs NGU, I would not have compared unscaled to upscaled!

If you want to compare NN vs NGU, we can do that very easily and directly in madVR without using a digicam.

which means we are at the fillrate problem again. but the digicam has other issues one of the mast of the ship nearly disappears on the NGU sharp digicam image (i guess the shutter time was to long the image is pretty bright in general) and this is not the case with the normal upscaled image. so it is not 100% faith full. the unscaled image get's some damage in this place too but it is not that extrem.

we could argue about that forever because it never really fair to compare to different image sizes. maybe with 2 CRT next to each other but i'm pretty sure we will still find issues in that setup.

Yes. Which is a good thing because it proves that my digicam isn't half bad. The photos faithfully show what my LCD monitor does, when displaying your DVD image unscaled vs upscaled. The photos show that displaying the content unscaled suffers more from the (sub)pixel grid of my LCD monitor than when using upscaling. The same effect exists when comparing 1080p content on a 1080p vs 4K display.

It's an important argument that displaying 1080p content untouched on a 1080p display doesn't faithfully represent the 1080p content, because many displays don't have perfectly rectangular pixels with 100% fillrate, and even more importantly, many displays have RGB subpixels that are positioned next to each other, so they don't cover the same area. This subpixel positioning can introduce additional aliasing/combing artifacts that aren't even in the source image, as seen in my digicam shots! So when showing a 1080p frame on a 1080p display, actually image quality gets lost due to the gaps between the pixels and the placement of the RGB subpixels. These problems of imperfect display technology are nicely reduced by upscaling the 1080p image to a higher resolution. And my digicam photos show that quite nicely.

it a RGB pixel type correction still on your to do list? (it has a specail name i didn't remember c... correction)

Yes, but I also used it for the AR corrected image (which was AR corrected by doing NGU Sharp + Grain + SSIM1D downscaling), so it's still a fair comparison.

u have to supersample the the image to add grain or madVR will use jinc.
supersampling isn't indicating it is doing it in one way only so we can't really call that image unscaled anymore.

and add grain clearly benefits from a higher resolution. with is a clear plus for your argument.

Yes, it's a trade. But the positives *FAR* outweigh the negatives, IMHO.

fine by me. i guess we are done with the topic now?