Dither stabilisation

https://forums.tigsource.com/index.php?topic=40832.msg1363742#msg1363742

by robin_reala • 9 years ago

291 20 9 years ago

forums.tigsource.com

mrob• 9 years ago

The dithered highlight on the sphere looks odd because it was calculated in a non-linear colorspace. You can get better looking results by converting to a linear colorspace, dithering, then converting back to sRGB. Eg, in ImageMagick, using the same 8x8 matrix:

convert Dither2-Exp_crop_resize.png -colorspace RGB -ordered-dither o8x8,2 -colorspace sRGB Dither2-Exp_dithered.png

Result: https://imgur.com/a/81bNg

klodolph• 9 years ago

Although in this case, the conversion back to sRGB is just the identity…

amluto• 9 years ago

True. But if the dither mask was computed without taking into account sRGB's nonlinearity (which I assume it was), then the algorithm will have the same incorrect output.

tizzdogg• 9 years ago

I replied on the original thread too, but I wonder if it’d work to use a mipmapped dither texture in texture space like this old hatching paper: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.28.... It probably won’t give you exact 1 pixel dither without some hacks, but maybe something along those lines could work.

bno1• 9 years ago

instead of using textures and downsampling them, would a mathematical formula for the dither pattern applied in the shader work best?

kmnt• 9 years ago

That is really cool, especially for the time. Only downside is I think some uv mapping is required to map the hatching onto the models. Here is a video of the paper https://www.youtube.com/watch?v=g6Ta35AiE1I

grenoire• 9 years ago

I think the problem now is that the dithering is no longer done on the screen space, but instead looks like a texture slapped on top of the objects in orthogonal projection. You can easily notice it in the comparison shots; you'll notice that the sphere projection is angled at around 20-30 degrees, and it's not pixel-perfect on the screen as the original effect is.

I'm not a fan, to be fair.

cutcss• 9 years ago

yeah looks strange; I think motion blur to reduce "discomfort" when rotating the camera may have been a better compromise; or maybe removing all single-out black pixels (e.g. those with no adjacent black pixels in a 4x4 radius).

rossy• 9 years ago

Yeah, I couldn't agree more. The original effect is really interesting and evokes nostalgia of old Mac games, as the developer intended. I think he's reacting too harshly to criticism, or overthinking things, because my reaction to the pixel swimming effect matched his first assumption: "yeah that's how dither works."

The screen-space solution for stable dithering looks fine, and I'm glad it will be an option in the final game, but the sphere mapping looks weird. With non-screen-aligned "pixels" it looks like it's just a normal 3D game but with a "dither" texture on every object, rather than a true 1-bit game. I just hope the aesthetic from the demo isn't watered down too much for the final game.

KozmoNau7• 9 years ago

The original effect is how it's supposed to look and how it would have looked on an old computer with 1-bit graphics.

If there is an option to choose between the different effects, I would personally go with the first one, because that's exactly how I remember old games looking.

teej• 9 years ago

This post is written by the creator of the wonderful game Papers, Please.

Franciscouzo• 9 years ago

I wonder how using error diffusion on the time axis would look, I guess it would stabilise the image a bit.

ttoinou• 9 years ago

It would look like sensor noise ?

thanatropism• 9 years ago

Noise has samples that are statistically independent from each other. Diffusion has samples such that their increments are independent. But it's path dependent.

contravariant• 9 years ago

Wouldn't error diffusion result in the opposite of stabilisation?

beeforpork• 9 years ago

What a delightful attention to detail! I love this!

It is totally not weird to me to spend time on something with the goal for it not to be noticed in the end.

monk_e_boy• 9 years ago

I couldn't find out why he wanted a 1 bit image. Was there a reason other than "it looks cool"?

jcl• 9 years ago

From the first post of the forum thread:

"My first computer was a Mac Plus. I've always had a nostalgia-softened spot in my heart for 1-bit graphics. I'd like to capture the detailed black & white look of old Mac games in a realtime 1st person game."

amai• 9 years ago

londons_explore• 9 years ago

This is the approach I'd use...

* Take the previous frame.

* Warp to the shape of the new frame using motion vectors.

* Dither the new frame using the warped old frame to modulate the white/black threshold when carrying forward the error to the next pixel.

That will lead to a pixel remaining white/black, and only a minimal number of pixels changing color between frames.

Dither stabilisation

Related Stories