madVR - high quality video renderer (GPU assisted)

[aufkrawall]

There's surely truth in this, but usually downscaling doesn't introduce many artifacts and C-R is neither blurry nor overly sharp.
Dunno if I'd call it "groundtruth" though.

[Ver Greeneyes]

Originally Posted by Thunderbolt8

my point is when watching a movie which needs to be upscaled it will not be downscaled first then. only upscaled. so the upscaled picture should be compared to the source image to clarify which algorythm comes closest to the source at this resolution.

How do you define 'the source at this resolution'? Do you downscale the upscaled image - if so, using which algorithm? Do you compare the upscale to the 'ideal' upscale using a more expensive algorithm like waifu2x? Do you compare it to the nearest neighbor upscale from a distance where you can no longer see the individual pixels? At some point you're going to have to make a judgment call. With madshi's way of downscaling first, at least you know what the source looks like. The only question is what impact the downscaling algorithm makes, and how it compares to how other sources are downscaled.

[Thunderbolt8]

the source at its native resolution I meant

Originally Posted by Ver Greeneyes

The only question is what impact the downscaling algorithm makes, and how it compares to how other sources are downscaled.

this point should definately be clarified before continuing with these comparisons

[Telion]

A no-brainer solution is to take the same screenshot from original DVD and BD sources of some initially analog video. E.g. one from Mononoke would be pretty good since this anime is known to be pure analog in production. So all dissimilarities between 2 reference screenshots would be almost completely due to a difference in scanning resolution from an analog film. Moreover, an upscale from SD to FHD would be much closer to a real-life scenario.

[Akeno]

Originally Posted by Telion

A no-brainer solution is to take the same screenshot from original DVD and BD sources of some initially analog video. E.g. one from Mononoke would be pretty good since this anime is known to be pure analog in production. So all dissimilarities between 2 reference screenshots would be almost completely due to a difference in scanning resolution from an analog film. Moreover, an upscale from SD to FHD would be much closer to a real-life scenario.

Were the DVD and Blu-Ray released at the same time? If not, we run into the issue of studios "touching up" the originals for Blu-Ray release as was the case of Tropic Thunder if I'm not mistaken. The studio applied bad sharpening and noise reduction that actually lowered detail compared to the DVD release.

[Thunderbolt8]

if possible, Im in favour of using upscale comparisons of 720p to 1080p and 1080p to 4k. that probably already is (720p to 1080p) and definately will be the most used cases of upscaling in the future and imho it makes sense to focus on these resolution when doing comparisons and not compare mini pictures.

[Ver Greeneyes]

The difficulty with 720p to 1080p is that it's not straight-up doubling, so not all algorithms can be applied directly. You either have to double the image using something like SuperXBR or NNEDI3, then downscale using another algorithm, or upscale using a scale-agnostic algorithm like Jinc.

However, if a movie is filmed in UltraHD, and they release UltraHD and FullHD versions, we'd be able to compare scaling algorithms on the downscaling algorithm that studios are using. The algorithm might still differ from studio to studio though, so it's not like there's One True Solution to this.

[Akeno]

Originally Posted by Ver Greeneyes

However, if a movie is filmed in UltraHD, and they release UltraHD and FullHD versions, we'd be able to compare scaling algorithms on the downscaling algorithm that studios are using. The algorithm might still differ from studio to studio though, so it's not like there's One True Solution to this.

In the end, it's basically what we are doing now. I doubt the studios scan the masters twice for DVD and Blu-Ray releases but it's anyone's guess on what they do.

[aufkrawall]

Originally Posted by SecurityBunny

Just out of curiosity, what is everyone currently using for debanding strength and dithering mode?

High deband & quality trade-off checkboxes unchecked, I really hate banding.
I find Error Diffusion totally useless with my 8 bit QX2710 bypass display, thus just ordered dithering.

Banding can be real-hardcore in real world content:
http://www18.zippyshare.com/v/0Jil84gy/file.html
madshi's deband filter is doing an excellent job.

[Telion]

Originally Posted by Akeno

I doubt the studios scan the masters twice for DVD and Blu-Ray releases but it's anyone's guess on what they do.

They do if BD is released long after the initial DVD release, as with Mononoke. Of course, the original source must be of high quality, as heavily remastered/retouched ones obviously should be avoided in such comparisons.

[DarkSpace]

Originally Posted by Akeno

In the end, it's basically what we are doing now. I doubt the studios scan the masters twice for DVD and Blu-Ray releases but it's anyone's guess on what they do.

Well, it's certainly not Anime, but they did for Star Trek (and did it really well, at that), but there it was advertised. Unfortunately, I don't remember enough to say that it was prominently advertised, but I believe it was.

Edit: Well, Telion mentioned the core information already some hours ago. I should go to bed now.

[Akeno]

Originally Posted by DarkSpace

Well, it's certainly not Anime, but they did for Star Trek (and did it really well, at that), but there it was advertised. Unfortunately, I don't remember enough to say that it was prominently advertised, but I believe it was.

Are we talking about the new Star Treks or the older movies?

[DarkSpace]

Originally Posted by Akeno

Are we talking about the new Star Treks or the older movies?

I'm talking about the series here... as far as I know, the movies I-X were rescanned only. I've got Star Trek TOS and TNG here, and I've heard that the animates series is also (going to be?) available on BD. Anyway, even TOS' "original effects" video has been remastered in so far as that the footage has been cleaned up after being re-scanned, so it's unsuitable for comparison, I'd guess.

I don't know whether its the NVidia 353.54 drivers or Windows 10 or madVR or MPC-HC, but in Windows 10, I am unable to use the seek bar in madVR exclusive mode. The mouse pointed can never reach the bottom of the screen and just resets itself to the middle of the screen. Anyone know which software is the issue?

Also, I noticed that latest NVidia Windows 10 drivers do not support CUDA. Is that a problem for using CUVID Decoding in LAV Filters or for madVR or anything else for that matter? Doesn't seem to affect games or anything else I use, but AFAIK CUDA is like the core architecture for NVidia, so..?

[aufkrawall]

Seekbar works fine here with same GPU and OS in D3D9 and 11, driver is 353.49.
Don't use drivers from WU like 353.54, they lack Cuda and/or OpenCL. Pure crap.

To prevent driver updates from WU for certain devices under Windows 10, see this:
https://support.microsoft.com/de-de/kb/3073930
If it doesn't work, try this script (no guarantee from my side for anything):
http://www68.zippyshare.com/v/wFdDMPMV/file.html

Originally Posted by aufkrawall

Seekbar works fine here with same GPU and OS in D3D9 and 11, driver is 353.49.
Don't use drivers from WU like 353.54, they lack Cuda and/or OpenCL. Pure crap.

To prevent driver updates from WU for certain devices under Windows 10, see this:
https://support.microsoft.com/de-de/kb/3073930
If it doesn't work, try this script (no guarantee from my side for anything):
http://www68.zippyshare.com/v/wFdDMPMV/file.html

Really? Then why is ED working with 353.54 drivers then?

[x7007]

NEEDI3 in Madvr 88.20 works in Potplayer + windows 10 + NVidia 353.54 + CUVID ?

Because my Rendering always 5.32ms and doesn't change when I change the madvr settings from 64 to 256 or any heavy setting.

and the bar works for me in Potplayer in exclusive mode + D3D11

[michkrol]

Originally Posted by XMonarchY

Really? Then why is ED working with 353.54 drivers then?

Because it uses Direct Compute?

Originally Posted by x7007

NEEDI3 in Madvr 88.20 works in Potplayer + windows 10 + NVidia 353.54 + CUVID ?

Because my Rendering always 5.32ms and doesn't change when I change the madvr settings from 64 to 256 or any heavy setting.

Check the OSD (Ctrl+J to toggle), you should get something like:
image > NNEDI3 ... or chroma > NNEDI3.
Then check with MPC-HC, just to be sure.
Then follow aufkrawall's advice from 2 posts up/back

[x7007]

Originally Posted by michkrol

Because it uses Direct Compute?



Check the OSD (Ctrl+J to toggle), you should get something like:
image > NNEDI3 ... or chroma > NNEDI3.
Then check with MPC-HC, just to be sure.
Then follow aufkrawall's advice from 2 posts up/back

For some weird reason it says Jinc AR even though I selected Chroma upscaling NEEDI3
All others works expect NEEDI3, if I select it , I see Chroma Jinc AR..

anyone has fix ?

Sometimes it's hard to go into exclusive mode in windows 10 , it tries to go in but flashes and you in black screen while the movie keep running, the only fast fix is to run the file again from the playlist, but it only happens randomly when you run the file, then it will always go exclusive without issue.

[huhn]

Originally Posted by x7007

anyone has fix ?

use drivers with openCL.