Image Gallery of madVR Image Processing Settings

[Warner306]

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 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
3D LUT Applied

tone map HDR using pixel shaders:

2,890 nits BT.2020 -> 150 nits BT.709 (480 target nits) (highlight recovery strength: high)

3,531 nits BT.2020 -> 150 nits BT.709 (480 target nits) (highlight recovery strength: high)

Mad Max Fury Road (HDR to SDR):

color tweaks for fire & explosions: disabled
color tweaks for fire & explosions: balanced
color tweaks for fire & explosions: high

Batman v Superman (HDR to SDR):

highlight recovery strength: none
highlight recovery strength: medium

X-Men Apocalypse (HDR to SDR):

dynamic clipping: disabled
dynamic clipping: enabled
no tone mapping: clipping

Star Wars The Last Jedi (HDR to SDR):

measure each frame: disabled
measure each frame: enabled

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

dynamic target nits: Disabled
dynamic target nits: Enabled

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

LG C8's Static Tone Mapping
LG C8's Static Tone Mapping + madVR's Dynamic Tone Mapping

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

LG C8's Static Tone Mapping
LG C8's Static Tone Mapping + madVR's Dynamic Tone Mapping
LG C8's Dynamic Tone Mapping On
LG C8's Dynamic Tone Mapping On + madVR's Dynamic Tone Mapping

chroma upscaling:

Comparison of Chroma Upscaling Algorithms

image downscaling:

Sunset - 4K -> 1080p:

DXVA2 (Nvidia)
Bicubic150 + LL + AR
SSIM 1D 100% + LL + AR
Jinc + LL + AR

image upscaling:

Clown:

Bilinear
Bicubic
Lanczos4
Jinc

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
Photoshop Downscaled 480p
Lanczos3 - no AR
Jinc + AR
super-xbr100 + AR
NGU Anti-Alias very high
NGU Standard very high
NGU Sharp very high

American Dad:

Original
Jinc + AR
super-xbr100 + AR
NNEDI3 256 neurons + SuperRes (4)
NGU Sharp very high

Wall of Books:

Original 480p
Lanczos3 - no AR
Jinc + AR
super-xbr100 + AR
NGU Anti-Alias very high
NGU Standard very high
NGU Sharp very high

Comic Book:

Original 1080p
Photoshop Downscaled 540p
Lanczos3 - no AR
Jinc + AR
super-xbr100 + AR
NGU Anti-Alias very high
NGU Standard very high
NGU Sharp very high

Corporate Photo:

Original 1080p
Photoshop Downscaled 540p
Lanczos3 - no AR
Jinc + AR
super-xbr100 + AR
NGU Anti-Alias very high
NGU Standard very high
NGU Sharp very high

Bilinear (Nvidia Shield upscaling algorithm)

image enhancements:

Saving Private Ryan:

Native Original
sharpen edges (4.0) + AR
crispen edges (3.0) + AR
LumaSharpen (1.50) + AR
AdaptiveSharpen (1.5) + LL + AR

upscaling refinement:

Original Castle Image (before 50% downscale)
Downscaled Castle Image resized using super-xbr100

super-xbr100 + FineSharp (4.0)
super-xbr100 + SuperRes (4)
super-xbr100 + AdaptiveSharpen (0.8)

reduce banding artifacts:

Demonstration of Debanding

1080p Blu-ray Credits:

Original
Debanding low
Debanding medium
Debanding high

reduce ringing artifacts:

Lighthouse Top:

No Deringing
madVR Deringing

DVD Animated:

No Deringing
madVR Deringing

reduce compression artifacts:

Animated:

NGU Sharp very high
NGU Sharp very high + RCA very high / strength:8

Music Video:

Original
RCA very high / strength:12

reduce random noise:

Saving Private Ryan:

Original
Denoising strength: 2
Denoising strength: 3
Denoising strength: 4
Denoising strength: 5

Lord of War:

Original
Denoising strength: 1
Denoising strength: 2
Denoising strength: 3
Denoising strength: 4

dithering:

Dithering to 2-bits:

2 bit Ordered Dithering
2 bit No Dithering

[Asmodian]

Nice collection, thanks!

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

[Warner306]

Originally Posted by Asmodian

Nice collection, thanks!

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

Sure, you might as well before this thread disappears.

[huhn]

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]

Originally Posted by huhn

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]

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.

[ashlar42]

Thanks Warner306. Much appreciated!

[Warner306]

Originally Posted by huhn

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]

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]

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]

Originally Posted by Warner306

Dithering to 2-bits:

2 bit Ordered Dithering
2 bit No Dithering

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.

[nevcairiel]

Indeed, the bit depth is per color.

[Warner306]

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]

Your conclusions match mine very well.

[foxyshadis]

Originally Posted by Warner306

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]

Originally Posted by foxyshadis

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]

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]

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.

[Warner306]

Originally Posted by huhn

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]

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