Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

mdapt dedithering doesn't blend very well with bigger grids #156

Open
whatwwkd4496 opened this issue Nov 28, 2019 · 7 comments
Open

mdapt dedithering doesn't blend very well with bigger grids #156

whatwwkd4496 opened this issue Nov 28, 2019 · 7 comments

Comments

@whatwwkd4496
Copy link

whatwwkd4496 commented Nov 28, 2019

Original
Original
mdapt
mdapt
gdapt
gdapt
cbod
cbod

infact none of the available shaders able to blend the dithering in this scene.
for reference purposes heres what it looks like with kawase0:

kawase0
kawase0

blur7
blur7
blur7 completely blends checkerboard pattern

@whatwwkd4496
Copy link
Author

whatwwkd4496 commented Nov 28, 2019

i've done some manual tinkering of this scene to get an idea on how it would turned out
manual tinkering
manual tinkering by splitting odd and even pixels verticaly and horizontaly and further splitting halved pixels to quarter size to get 4 split images then blend them into two final look. and the bottom right is the combination of two final images to get one combined final-final look.

@hizzlekizzle
Copy link
Collaborator

yeah, mdapt is pretty cautious about dithering detection. gdapt is from the same author as mdapt and is designed to catch more patterns at the cost of more false-positives.

GTUv50 is another good shader for handling dithering. You can drop the horizontal/vertical signal bandwidth until the dithering starts to blend. It's basically a blur that you can fine-tune:
image

@whatwwkd4496
Copy link
Author

could you make an edge preserving dedithering with two variants, one is for graphics priority with more aggressive blending with more errors, and another is for text priority with subtle blending without causing text errors. so two needs can be satisfied. i'm guessing the text priority part is already exist and fullfilled its purpose. because fighting games like this could benefit more from graphics than text.

@hizzlekizzle
Copy link
Collaborator

It's possible, but it's not just text that gets blended with false positives; any small detail like eyes can get averaged away.

@whatwwkd4496
Copy link
Author

yes please.

@whatwwkd4496
Copy link
Author

i've been thinking. this dithering is basicly checkerboard rendering of two layer of frames that overlays on top of each other to produce transparency effect, so it's two half-pixel-frames that needs to be upscaled to full frames and overlaid on top of each. so we already have upscaling filters like scalex, xbr, and nnedi3. first we need to upscale odd and even pixels separately then blend them to a single frame. is anyone willing to attempt that approach?

@whatwwkd4496
Copy link
Author

i found something good
Eternal Champions (USA)-210627-205759-1x_DitherDeleterV3-Smooth- 32 _115000_G_4x
but its an ESRGAN model (DitherDeleterV3) and rendering time is terrible.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants