Image Gallery of madVR Image Processing Settings [Archive]

View Full Version : Image Gallery of madVR Image Processing Settings

Warner306

1st May 2018, 20:22

Below is a quick introduction to madVR's various image processing settings.

Some images were originally posted by madshi. The remaining were provided by me. Thanks to quietvoid for the Spears & Munsil Demo Footage images, huhn for the "DVD Animated" deringing submission, Neo-XP for the color tweaks for fire & explosions and highlight recovery images, and Soulnight for the dynamic clipping images.

This tool (https://www.diffchecker.com/image-diff) can be used to highlight the parts of the image that have been changed.

The link name will appear on the browser tab and at the bottom of the image.

3D LUT gamut correction:

Original - No 3D LUT (http://upload.vstanced.com/image/A35T)
3D LUT Applied (http://upload.vstanced.com/image/A35Z)

tone map HDR using pixel shaders:

2,890 nits BT.2020 -> 150 nits BT.709 (480 target nits) (highlight recovery strength: high) (http://upload.vstanced.com/images/2019/12/07/6e9311eb9c06249b9e46e7e397abc98b.png)

3,531 nits BT.2020 -> 150 nits BT.709 (480 target nits) (highlight recovery strength: high) (http://upload.vstanced.com/images/2019/12/07/8c9468de288d286dae09d60ec88f4331.png)

Mad Max Fury Road (HDR to SDR):

color tweaks for fire & explosions: disabled (http://upload.vstanced.com/image/AC57)
color tweaks for fire & explosions: balanced (http://upload.vstanced.com/image/AC5f)
color tweaks for fire & explosions: high (http://upload.vstanced.com/image/AC5h)

Batman v Superman (HDR to SDR):

highlight recovery strength: none (http://upload.vstanced.com/image/A3GJ)
highlight recovery strength: medium (http://upload.vstanced.com/image/A3Gx)

X-Men Apocalypse (HDR to SDR):

dynamic clipping: disabled (http://upload.vstanced.com/image/ACoR)
dynamic clipping: enabled (http://upload.vstanced.com/image/ACoH)
no tone mapping: clipping (http://upload.vstanced.com/image/ACAn)

Star Wars The Last Jedi (HDR to SDR):

measure each frame: disabled (http://upload.vstanced.com/image/AC4K)
measure each frame: enabled (http://upload.vstanced.com/image/AC4N)

Spears & Munsil UHD HDR Benchmark 10,000 nits MaxCLL (HDR to SDR):

dynamic target nits: Disabled (http://upload.vstanced.com/image/AFJS)
dynamic target nits: Enabled (http://upload.vstanced.com/image/AFJ4)

Spears & Munsil UHD HDR Benchmark 10,000 nits MaxCLL (HDR10 Passthrough):

LG C8's Static Tone Mapping (http://upload.vstanced.com/image/AFMI)
LG C8's Static Tone Mapping + madVR's Dynamic Tone Mapping (http://upload.vstanced.com/image/AFMO)

Spears & Munsil UHD HDR Benchmark 10,000 nits MaxCLL (HDR10 Passthrough):

LG C8's Static Tone Mapping (http://upload.vstanced.com/image/AFMT)
LG C8's Static Tone Mapping + madVR's Dynamic Tone Mapping (http://upload.vstanced.com/image/AFMw)
LG C8's Dynamic Tone Mapping On (http://upload.vstanced.com/image/AFMr)
LG C8's Dynamic Tone Mapping On + madVR's Dynamic Tone Mapping (http://upload.vstanced.com/image/AFMZ)

chroma upscaling:

Comparison of Chroma Upscaling Algorithms (https://forum.doom9.org/showthread.php?p=1817934#post1817934)

image downscaling:

Sunset - 4K -> 1080p:

DXVA2 (Nvidia) (http://upload.vstanced.com/image/A39H)
Bicubic150 + LL + AR (http://upload.vstanced.com/image/A393)
SSIM 1D 100% + LL + AR (http://upload.vstanced.com/image/A39D)
Jinc + LL + AR (http://upload.vstanced.com/image/A39q)

image upscaling:

Clown:

Bilinear (http://madvr.com/doom9/clown/clownBilinear.png)
Bicubic (http://madvr.com/doom9/clown/clownBicubic.png)
Lanczos4 (http://madvr.com/doom9/clown/clownLanczos4.png)
Jinc (http://madvr.com/doom9/clown/clownJinc.png)

image doubling:

Note on image doubling: The "Original 1080p" images in the comparisons below can make for a difficult reference because Photoshop tends to alter image detail significantly when downscaling. The color is also a little different. These images are still available as a reference as to how sharp the upscaled image should appear.

Video Game Poster:

Original 1080p (http://upload.vstanced.com/image/A3U1)
Photoshop Downscaled 480p (http://upload.vstanced.com/image/A3XB)
Lanczos3 - no AR (http://upload.vstanced.com/image/A3Xg)
Jinc + AR (http://upload.vstanced.com/image/A3XQ)
super-xbr100 + AR (http://upload.vstanced.com/image/A3XP)
NGU Anti-Alias very high (http://upload.vstanced.com/image/A3XF)
NGU Standard very high (http://upload.vstanced.com/image/A3X9)
NGU Sharp very high (http://upload.vstanced.com/image/A3X5)

American Dad:

Original (http://madshi.net/madVR/AmericanDad/AmericanDadOrg.png)
Jinc + AR (http://madshi.net/madVR/AmericanDad/AmericanDadJinc.png)
super-xbr100 + AR (http://madshi.net/madVR/AmericanDad/AmericanDadSuperXbr.png)
NNEDI3 256 neurons + SuperRes (4) (http://madshi.net/madVR/AmericanDad/AmericanDadNNEDI256SR4.png)
NGU Sharp very high (http://madshi.net/madVR/AmericanDad/AmericanDadNG1.png)

Wall of Books:

Original 480p (http://upload.vstanced.com/image/A3XU)
Lanczos3 - no AR (http://upload.vstanced.com/image/A3X3)
Jinc + AR (http://upload.vstanced.com/image/A3XH)
super-xbr100 + AR (http://upload.vstanced.com/image/A3XC)
NGU Anti-Alias very high (http://upload.vstanced.com/image/A3XR)
NGU Standard very high (http://upload.vstanced.com/image/A3X4)
NGU Sharp very high (http://upload.vstanced.com/image/A3XS)

Comic Book:

Original 1080p (http://upload.vstanced.com/image/A3Us)
Photoshop Downscaled 540p (http://upload.vstanced.com/image/A3P2)
Lanczos3 - no AR (http://upload.vstanced.com/image/A3Pt)
Jinc + AR (http://upload.vstanced.com/image/A3Pa)
super-xbr100 + AR (http://upload.vstanced.com/image/A3Ps)
NGU Anti-Alias very high (http://upload.vstanced.com/image/A3PM)
NGU Standard very high (http://upload.vstanced.com/image/A3P1)
NGU Sharp very high (http://upload.vstanced.com/image/A3Pc)

Corporate Photo:

Original 1080p (http://upload.vstanced.com/image/A3Uc)
Photoshop Downscaled 540p (http://upload.vstanced.com/image/A3Xi)
Lanczos3 - no AR (http://upload.vstanced.com/image/A3Uy)
Jinc + AR (http://upload.vstanced.com/image/A3UJ)
super-xbr100 + AR (http://upload.vstanced.com/image/A3Xq)
NGU Anti-Alias very high (http://upload.vstanced.com/image/A3Xo)
NGU Standard very high (http://upload.vstanced.com/image/A3XD)
NGU Sharp very high (http://upload.vstanced.com/image/A3XA)

Bilinear (Nvidia Shield upscaling algorithm) (http://upload.vstanced.com/image/A3Ue)

image enhancements:

Saving Private Ryan:

Native Original (http://upload.vstanced.com/image/ASKx)
sharpen edges (4.0) + AR (http://upload.vstanced.com/image/ASKj)
crispen edges (3.0) + AR (http://upload.vstanced.com/image/ASKm)
LumaSharpen (1.50) + AR (http://upload.vstanced.com/image/ASKu)
AdaptiveSharpen (1.5) + LL + AR (http://upload.vstanced.com/image/ASKf)

upscaling refinement:

Original Castle Image (before 50% downscale) (http://madvr.com/doom9/castle/CastleBig.png)
Downscaled Castle Image resized using super-xbr100 (http://madvr.com/doom9/castle/ScalingAlgos/CastleSuperXbr.png)

super-xbr100 + FineSharp (4.0) (http://madvr.com/doom9/castle/Sharpening/CastleSuperXbrFineSharp.png)
super-xbr100 + SuperRes (4) (http://madvr.com/doom9/castle/Sharpening/CastleSuperXbrHqSuperRes.png)
super-xbr100 + AdaptiveSharpen (0.8) (http://madvr.com/doom9/castle/Sharpening/CastleSuperXbrAdaptiveSharpen.png)

reduce banding artifacts:

Demonstration of Debanding (http://madshi.net/madVR/debandComparison.png)

1080p Blu-ray Credits:

Original (http://upload.vstanced.com/image/ASwL)
Debanding low (http://upload.vstanced.com/image/ASw2)
Debanding medium (http://upload.vstanced.com/image/ASwd)
Debanding high (http://upload.vstanced.com/image/ASwE)

reduce ringing artifacts:

Lighthouse Top:

No Deringing (http://madvr.com/doom9/dering/lightHouseNoDering.png)
madVR Deringing (http://madvr.com/doom9/dering/lightHouse9019.png)

DVD Animated:

No Deringing (http://upload.vstanced.com/images/2018/09/12/99cfb5e23912b5309002626b9c2011ab.png)
madVR Deringing (http://upload.vstanced.com/images/2018/09/12/a9d4fe409e683d8858ec1d71d65b2dfd.png)

reduce compression artifacts:

Animated:

NGU Sharp very high (http://upload.vstanced.com/image/A39G)
NGU Sharp very high + RCA very high / strength:8 (http://upload.vstanced.com/image/A39F)

Music Video:

Original (http://upload.vstanced.com/image/AClM)
RCA very high / strength:12 (http://upload.vstanced.com/image/AClc)

reduce random noise:

Saving Private Ryan:

Original (http://upload.vstanced.com/image/ASKr)
Denoising strength: 2 (http://upload.vstanced.com/image/ASKw)
Denoising strength: 3 (http://upload.vstanced.com/image/ASK0)
Denoising strength: 4 (http://upload.vstanced.com/image/ASKk)
Denoising strength: 5 (http://upload.vstanced.com/image/ASK7)

Lord of War:

Original (http://upload.vstanced.com/image/ACYa)
Denoising strength: 1 (http://upload.vstanced.com/image/ACYc)
Denoising strength: 2 (http://upload.vstanced.com/image/ACYs)
Denoising strength: 3 (http://upload.vstanced.com/image/ACY1)
Denoising strength: 4 (http://upload.vstanced.com/image/ACYE)

dithering:

Dithering to 2-bits:

2 bit Ordered Dithering (http://madshi.net/2bitOrderedDithering.png)
2 bit No Dithering (http://madshi.net/2bitNoDithering.png)

Asmodian

1st May 2018, 21:17

Nice collection, thanks!

You do not mind if I put a link to this in my Options Explained post?

:thanks:

Warner306

2nd May 2018, 01:03

Nice collection, thanks!

You do not mind if I put a link to this in my Options Explained post?

:thanks:

Sure, you might as well before this thread disappears.

huhn

2nd May 2018, 02:03

are the deringing images correct?

i mean ringing is the last problem i have with these images and it's not really effect on them in this example.

Warner306

2nd May 2018, 08:06

are the deringing images correct?

i mean ringing is the last problem i have with these images and it's not really effect on them in this example.

I can see ringing, especially the first image of the lighthouse. It is pretty blatant ringing, to me. Maybe it's your monitor. That ringing is removed. The second image shows ringing on the blond woman's head and on the banister in the back. This ringing is also removed.

It must be hard to find images with obvious ringing that aren't compression artifacts. I just tried to upscale an image with Lanczos4 and couldn't get enough ringing to make a comparison photo. So I don't have any better examples to provide. I've removed edge enhancement in Blu-rays before, but this effect is not that obvious in test photos.

I did add a new set of images for reduce banding artifacts.

If anyone has better images, I would upload and add them. There is still room for a few more sets.

huhn

2nd May 2018, 10:33

ashlar42

2nd May 2018, 11:45

Thanks Warner306. Much appreciated!

Warner306

2nd May 2018, 16:52

i'm talking about the animated on. and as i see these are NN upscaled by a huge factor to make it more visible. my screen has nothing todo with the problem.

what is wrong with removing compression artefact ringing?
i mean that's the point of the reduce compression artifacts anti ringing.

here a quick and dirty alternative:
dvd -> dringing 2x NGU AA high (using he screenshoot function of madVR)
https://abload.de/img/new2deringingzgu2n.png
https://abload.de/img/new2ringingedue8.png

just the best i could do in a short time.

on the other animated images the first thing i see is aliasing and don't really care about the rest because deringing doesn't really help this image enough to shine.
using deringing on DVDs with ringing is huge and show a giganatic difference on a UHD screen.
far far from perfect because there are a lot of other issues in these images.

I honestly have trouble seeing the ringing in that photo. You have to understand something gets lost in translation when viewed on the Web and varies depending on the monitor you have. I had that problem when uploading the most recent debanding photos. In Photoshop on my TV, the difference is obvious. I tried an uncompressed TIFF and PNG, and it still looks different when viewed through Google Drive. I also made the mistake of compressing the image doubling photos. I'm not sure if I should try again or if the image doubling photos are good enough.

Ok, I can see the lines change, but the original ringing is not obvious enough. Like I said, it probably looks obvious on your end.

huhn

2nd May 2018, 19:53

my image are not compressed and the ringing is so bad it doesn't even fix it.

my TV/screens are calibrated there is no big difference between my different screens.

my point is that there so much aliasing that ringing is not jumping into your eyes because the image has clearly bigger issues.

if you want to see the difference here you go: https://www.diffchecker.com/image-diff

Warner306

2nd May 2018, 23:52

After looking at your images for a second time, I can see what you're talking about. I added the images to the first post. I also added another set of images I had to image doubling.

ashlar42

3rd May 2018, 13:43

Dithering to 2-bits:

2 bit Ordered Dithering (http://madshi.net/2bitOrderedDithering.png)
2 bit No Dithering (http://madshi.net/2bitNoDithering.png)

Note: Not all images display obvious differences between algorithms. The ones shown do. That's just the nature of how each algorithm works. The same would apply to actual content.
This seems as good a place as any to ask this question that I've wanted to ask for quite a while.

1 bit color should be black and white, correct?
2 bit color should offer 4 colors, correct?

Then why is it that when using 2 bit dithering I always count more than 4 colors? In your Avatar example, with dithering activated, the picture displays 37 individual colors (calculated in Photoshop, Mode Indexed Color).

Edit: unless it's 4^3 because the value applies to single RGB components. In that case 2 bit dithering allows 64 colors. Yup, that's it. In fact with 8 bit color you have a 255^3 colors palette (the usual 16M colors).
Well... at least your post made my brain go back to work. :D

nevcairiel

3rd May 2018, 16:18

Indeed, the bit depth is per color.

Warner306

5th May 2018, 00:59

Added a few more images to make for a more complete gallery.

I didn't provide any introduction or commentary, so I thought I'd do that now.

As someone who has viewed a number of test images with various settings, I can conclude the following. And remember, this is only my opinion...

When it comes to image scaling, what separates NGU from Lanczos, Jinc and super-xbr is its ability to find small image detail. The effect is similar to a focus on a camera or a deblurring filter in Photoshop. Almost all images I've seen show a difference, but sometimes you need to look at the image as a whole to appreciate it, rather than a few areas that change significantly where other scalers have made errors or interpreted things differently. The impact of this focus effect would increase with larger upscales and with images shown on larger screens like a projector. Much of the time, you may not be blown away by image scaling, but NGU is consistently better and always a little more precise and detailed. How much this matters depends on the scaling factor, scene being rendered and the display size.

Artifact removal can be inconsistent. Some artifacts are removed completely; others are only partially removed. And the strength needed to remove the artifact varies from a mid-range to a high-range value. In general, if you want to do more than blur or alter an artifact and remove it, you will have to use a mid to high value and run the risk of removing some image detail. This detail loss can be random or completely non-existent, depending on the image being corrected. reduce compression artifacts doesn't seem to be well-suited for large macroblocking. It is much better at correcting the type of noise and artifacts found in something like an iTunes download.

The majority of the other settings like image sharpening (e.g. image enhancements and upscaling refinement) are more predictable. I'm not sure if there is any value in Error Diffusion dithering over Ordered Dithering without very specific test patterns. At 10-bits, Error Diffusion dithering would be pointless.

I hope that doesn't ruffle any feathers, providing my opinion.

Asmodian

5th May 2018, 02:25

Your conclusions match mine very well. :)

foxyshadis

5th May 2018, 12:17

Sure, you might as well before this thread disappears.

I'd rather sticky it. So many people ask what the difference is, and there's so few references.

Warner306

5th May 2018, 16:10

I'd rather sticky it. So many people ask what the difference is, and there's so few references.

Ok, but I don't think my examples are very professional or high-quality. This thread can be accessed by a couple of links in other threads/forums but not for anyone casually browsing Doom9.

A sticky would be fine as the links are hosted by a free provider so they should never die. I'm just not sure if I'm the best person to provide image samples.

Warner306

6th May 2018, 18:27

Made a quick substitution for two image sets that made NGU look bad. I might have otherwise had a guilty conscience if this thread was stickied without madshi's approval.

huhn

7th May 2018, 11:02

Warner306

7th May 2018, 16:45

i highly value the replacements with better alternatives.

what makes a good example image in my point of view.

it doesn't show other major artifacts. so it doesn't distract from the shown filter in question.

upscaling an image with a high quality image scaler is fine in my view as long as this is communicated properly.

an image should have a resolution of 1080p before cropping or higher. maybe a 1080p and a 2160p version should be provided.
providing an 480p image with an filter on it will not help an user on a UHD screen.

cropping should be fine and used if it hides other artifacts on images.

if an compression reduction filter is used it should be used on an image where most if not all of this artifact in question is removed.

for this thread is not super important but an unscaled version should be provided somewhere so people can reproduce the results this is not really easy to do with anamorphic sources and is not really supportted by madVR screenshoot function.

I can agree with all of this. Creating a sticky for this thread does seem contrary to its original purpose. I was just showing the differences between algorithms, not trying to provide real world examples.

I don't have a 4K display available to me. Otherwise, I would scale everything from 1080p -> 4K. The best upscales seem to be slightly larger than 2x, so I went with 2.25x. Some of the images have artifacts because my only source for images is Google Images. I did discover that downscaling a 1080p image in Photoshop is better than a lower resolution image off a website. But it still displays some artifacts after downscaling, so I can't find perfect examples. I could come up with several more images but not perfect examples. You simply can't remove compression artifacts reliably in Photoshop. It has weak filters for this purpose.

It should be up to madshi if he wants to create an image gallery. He has never done this in the past, so maybe it's just not his thing. In that case, this thread should probably just be left as is.

I can't say that any of the algorithms is perfect, either. They all show some flaws no matter what I'm doing with the image. I still can't complain overall. But you can't say madVR is perfect.

Dorohedoro

7th May 2018, 18:31

Are you using the anti ringing filter on the Jinc captures?

Warner306

7th May 2018, 23:18

Are you using the anti ringing filter on the Jinc captures?

Yes, it is Jinc + AR. I'm sure the source has ringing and Jinc is simply not correcting this.

Only Lanczos was used without AR to show what most low-powered streaming boxes use. Most actually rely on Bilinear or Bicubic with only Kodi LibreELEC using Lanczos, all without anti-ringing. Some of those low-powered boxes actually score Nearest Neighbor on upscaling tests, even though they are likely using Bilinear.

huhn

7th May 2018, 23:50

the anti ringing in upscaler is not there to remove ringing it is used to stop image scaler from adding more.

foxyshadis

10th May 2018, 07:13

I can agree with all of this. Creating a sticky for this thread does seem contrary to its original purpose. I was just showing the differences between algorithms, not trying to provide real world examples.

I don't have a 4K display available to me. Otherwise, I would scale everything from 1080p -> 4K. The best upscales seem to be slightly larger than 2x, so I went with 2.25x. Some of the images have artifacts because my only source for images is Google Images. I did discover that downscaling a 1080p image in Photoshop is better than a lower resolution image off a website. But it still displays some artifacts after downscaling, so I can't find perfect examples. I could come up with several more images but not perfect examples. You simply can't remove compression artifacts reliably in Photoshop. It has weak filters for this purpose.

It should be up to madshi if he wants to create an image gallery. He has never done this in the past, so maybe it's just not his thing. In that case, this thread should probably just be left as is.

I can't say that any of the algorithms is perfect, either. They all show some flaws no matter what I'm doing with the image. I still can't complain overall. But you can't say madVR is perfect.

If you feel that strongly about it, no worries, I was mostly trying to communicate that I'm not going to delete anything and I think you did some really valuable work, trying to answer everyone's big question, "which resizer do I pick?" In a sense it's unanswerable, but with real-life examples someone can say "oh, I like that one" and go on with their day. To people who overthink things, like me, that's the critical difference between setting up madvr taking five minutes or five hours.

Warner306

10th May 2018, 16:37

If you feel that strongly about it, no worries, I was mostly trying to communicate that I'm not going to delete anything and I think you did some really valuable work, trying to answer everyone's big question, "which resizer do I pick?" In a sense it's unanswerable, but with real-life examples someone can say "oh, I like that one" and go on with their day. To people who overthink things, like me, that's the critical difference between setting up madvr taking five minutes or five hours.

I don't feel that strongly about it anymore. I don't think I'd like to see the thread disappear after taking about 30 edits to get all of the links in. I had to change image hosts three times after having issues with Google Drive sharing my email, having it downgrade links after a set amount of days and having another provider downgrade resolution and file size. Now that all of it is here, I may go back and improve a few sections if I feel motivated in the future.

zaemon

20th May 2018, 09:21

Great post.

ccrys

27th September 2018, 13:26

It would be more interesting at least for me if you post some "real life movies" not pictures , I mean ss's comparison for 1080p movies upscaled to 4k. Except some upscale refinent the differences are insignifiant, least for me :).

Warner306

27th September 2018, 16:08

I don't have a 4K display. You would still obviously see a difference. If you go from 1080p -> 4K, you have 4x as many luma pixels and chroma is upscaled 4x. All of those pixels will result in a different image because they have to be invented by the media player. It simply isn't possible for the image to be the same at 1080p and 4K because it isn't the same image. Have you ever played video games at 1080p vs. 4K? The only goal of image upscaling is to make the 1080p image look like a genuine, high resolution 4K image. If want to change the look of the image, you need to use other settings.

If you spend some time watching content, you can actually spot the a specific look provided from different settings (upscaling, sharpening, debanding, etc.). Those images are just there to demonstrate the differences between each settings choice on a computer monitor. I actually have a few more images to substitute. You posted three different places about the same topic. You should take the advice provided at AVSForums.

Warner306

27th September 2018, 16:32

Here: I found a few comparisons between 1080p upscaled to 4K with NGU Sharp very high versus native 4K. This is what you seem to be complaining about:

http://archimago.blogspot.com/2018/01/1080p-blu-ray-vs-4k-uhd-blu-ray-does.html

http://archimago.blogspot.com/2018/02/1080p-blu-ray-vs-4k-uhd-blu-ray-pacific.html

http://archimago.blogspot.com/2018/02/1080p-blu-ray-vs-4k-uhd-blu-ray.html

If you search for "Javs" at AVSForums, he has posted some better comparison screenshots in the past, but I can't find them through Google.

DTL

5th February 2021, 22:49

"The "Original 1080p" images in the comparisons below can make for a difficult reference because Photoshop tends to alter image detail significantly when downscaling."

I think Photoshop is completely wrong software for moving pictures production (downscaling). At least because it is software for static pictures processing. Being used for creating moving picture digital content it highly likely will made dataset with bad 'conditioning' for moving picture upsamplers caused significant ringing and alising for static frames and more badly flickering in dynamic frames sequence.

For sad real world in 2021 without still reference ISO/IEC/MPEG algoriphms on digital moving pictures production and for real moving pictures viewers community I think the only real source data for upscalers test is film transfers of some good production firms and also software render of also best industry leaders for moving pictures production like Pixar and may be some other.
There may be also some hand-made production downscalers for moving pictures - but this is far from real world of commertial moving pictures data content and also video-rip content made by amateurs using video-rip software.

So for testing upsamplers for displaying commetrial moving pictures content we only can select some frames from this content (also studios may change settings and software versions/algoriphms for rendering at different movie titles at different years of production).

For testing of upsample of video-ripper low res rips I think typical downsizer is something like Lanczos3 or may be even Bicubic. Presize Bicubic for prefiltering from VirtualDub. Also the source image for downsizing must be image from moving picture dataset like goog FullHD film scan/transfer or some good adjusted broadcast video cam of FHD/4K resolution.

The only goal of image upscaling is to make the 1080p image look like a genuine, high resolution 4K image.

The provided to consumer moving picture dataset is actually sort of 'compressed image data' and it must be upscaled (in theory to infinite digital samples size) before viewing. Output of MPEG decoder is also still 'compressed' to finite number of samples (usually even different number of samples for chroma and luma). So to be displayed properly it must be upscaled to the high enough ratio (at least 2x for luma and 4x for chroma). For moving picture data the upscaler must provide good result not only at static frame but with 'moving' frame sequence so displayed 'moving' objects do not change its size and shape because of upsampler's algoriphms.

Ofcourse in best case we need to use 'complementary' pair of compressor (downsizer) at production industry side and expander (upsizer) at consumer/viewer side. But in sad real world we can only test and look which upsizer is acceptable or best for each moving picture content.

Got EBU tech 3325 test pattern 7 for testing scalers of broadcast control monitors and made test with madVR 2x upscalers and some separate upscalers from Avisynth plugins.

Results combined to 1 image file: https://i1.imageban.ru/out/2021/02/06/51daae063fda26b70933ff03581cfc0a.png
https://i1.imageban.ru/thumbs/2021.02.06/51daae063fda26b70933ff03581cfc0a.png (https://imageban.ru/show/2021/02/06/51daae063fda26b70933ff03581cfc0a/png)

In compare with some slow in processing with high number of taps sinc (1D+1D V+H separated processed with sinc kernel) and jinc (2D/planar processing with jinc kernel) it looks EBU engineers live in separate perfect world and not very good connected to real sad world of digital moving pictures content production companies. So most of build-in madVR upsamplers works bad enough on EBU tech sample dataset and mostly designed to fight against ringing on real world badly 'conditioned' digital moving pictures data while trying to keep sharpness as great as possible.

The most useful practical observation: some madVR resizers DO work better with EBU tect 3325 test pattern when scaling in sigmodial light providing less alising on high diagonal spatial frequencies and some other produces significally worse result. So it is important to use correct options of resizer + sigmodial light option at least at digital moving pictures content 'conditioned' for EBU control monitors.