Higher quality lightmap sampling #16740
Conversation
JMS55
added
the
S-Needs-Review
|
imo this should be optional (with at least linear being the other option). Try with a low res light map to see the effect. |
There was a problem hiding this comment.
Looks good modulo a little inaccuracy.
I could go either way on exposing an option for bilinear, and I won't block the PR on that. A quote from the Bakery author's blog post: https://ndotl.wordpress.com/2018/08/29/baking-artifact-free-lightmaps/
Bicubic interpolation. If you are not shipping on mobile, there are exactly 0 reasons to not use bicubic interpolation for lightmaps. Many UE3 games in the past did that, and it is a great trick. But some engines (Unity, I’m looking at you) still think they can get away with a single bilinear tap in 2018. Bicubic hides low resolution and makes jagged lines appear smooth. Sometimes I see people fixing jagged sharp shadows by super-sampling during bake, but it feels like a waste of lightmapping time to me.
See the post for a screenshot comparison.
The "if you are not shipping on mobile" comment is 6 years old, so it may well just be time to say bicubic everywhere.
| let p0 = (vec2(iuv.x + h0x, iuv.y + h0y) - 0.5) * texel_size; | ||
| let p1 = (vec2(iuv.x + h1x, iuv.y + h0y) - 0.5) * texel_size; | ||
| let p2 = (vec2(iuv.x + h0x, iuv.y + h1y) - 0.5) * texel_size; | ||
| let p3 = (vec2(iuv.x + h1x, iuv.y + h1y) - 0.5) * texel_size; |
There was a problem hiding this comment.
nit: Odd indentation here
There was a problem hiding this comment.
Copy+pasted tabs :/. Fixed.
| } | ||
|
|
||
| fn h1_approx(a: f32) -> f32 { | ||
| return 1.0 + a * (0.24 * a - 0.04); |
There was a problem hiding this comment.
I saw that comment, but the rounding on the coefficients for h1_approx leads to some serious error:
I calculated out the real coefficients here:
There was a problem hiding this comment.
| return 1.0 + a * (0.24 * a - 0.04); | |
| return 1.00158 + a * (0.230963 * a - 0.0353001); |
There was a problem hiding this comment.
I plugged it into desmos. Both yours and my h1 approx have inaccuracies. The difference is whether it occurs before or after the midpoint.
There was a problem hiding this comment.
Oops, I replaced the - with a +! Sign error. Ignore my comment, your code is fine the way it is
| } | ||
|
|
||
| fn h0_approx(a: f32) -> f32 { | ||
| return 0.2 + a * (0.24 * a - 0.44); |
There was a problem hiding this comment.
For comparison with the code below, this is a fine approximation.
|
Based on these images from @pcwalton I don't think bicubic is correct: https://discord.com/channels/691052431525675048/743663924229963868/1316259592308527134
|
Objective
Solution
Testing
Changelog