Hacker News new | ask | show | jobs
by grgrdvrt 1913 days ago
Thank you, it doesn't sound negative at all, in fact you raise interesting points!

the "obviousness" of the connections is tricky. The blending is in the linear RGB space. I initially tried sRGB but it was terrible. I'm not sure how to improve it. At the moment the interpolation is done through the varyings between the vertex and fragment shader. This is quite limiting.

About the lack of deliberate choice, in theory you can reach any rgb color. In practice this is limited by the initial colors I chose (they are smoother than pure RGB-CMY). I'm not sure how much of the RGB space I'm missing.

For the antialiasing issue I'll look into it. I'm not sure I correctly set the AA in webgl. Could also be due to some issue with the size of the canvas

Thank you again!
1 comments
raphlinus 1913 days ago
Interpolate in a perceptual space such as Oklab?

Possibly use a radial basis function for interpolation rather than triangles? I don't have JavaScript code handy but rbf-interp[1] is Rust code that could either be adapted or called from wasm.

[1]: https://crates.io/crates/rbf-interp

link
grgrdvrt 1913 days ago
The triangles-based interpolation is indeed problematic. It has the advantage to be fast and super easy to implement in webgl with an arbitrary number of points. I don't understand everything yet about the rbf interpolation but it looks very interesting, I'll look into it.

A better color space can also help with some issues (in the demo the darker colors are harder to reach when blending with the black point than lighter colors are when the white point is involved) but how I understand it the colors in a triangle are influenced by the colors and shapes of the surrounding triangles, whatever the color space.

I must say that I'm very happy to have had your input on this. I'm a big fan of your work and your blog is a real treasure!

Thank you!

link
nom 1913 days ago
This! Interpolating in linear RGB space is the biggest issue, I'm sure it will improve a lot simply by using a more suitable colorspace. Triangles might not be noticable at all then, except in extreme cases (which are undesirable anyways).
link
grgrdvrt 1913 days ago
I'll try a different color space, for sure, but I think the interpolation method is the real culprit. I have the feeling that in a configuration like this:

  A-B
  |/|
  C-D
the position of D could have an impact on how ABC is rendered.

But I'll try another color space first, it should be easier to implement.

link