I am trying to implement a texture cubic projection inside my WebGL shader, like in the picture below:
What I tried so far:
I am passing the bounding box of my object (the box in the middle of the picture) as follows:
uniform vec3 u_bbmin;
uniform vec3 u_bbmax;
... so the eight vertexes of my projection box are:
vec3 v1 = vec3(u_bbmin.x, u_bbmin.y, u_bbmin.z);
vec3 v2 = vec3(u_bbmax.x, u_bbmin.y, u_bbmin.z);
vec3 v3 = vec3(u_bbmin.x, u_bbmax.y, u_bbmin.z);
...other combinations
vec3 v8 = vec3(u_bbmax.x, u_bbmax.y, u_bbmax.z);
At the end, to sample from my texture I need a map in the form of:
varying vec3 v_modelPos;
...
uniform sampler2D s_texture;
vec2 tCoords = vec2(0.0);
tCoords.s = s(x,y,z)
tCoords.t = t(y,y,z)
vec4 color = texture2D(s_texture, tCoords);
I was able to implement spherical and cylindrical projections, but I am stuck now how to get this kind of cubic map, The texture shall stretch to the whole bounding box, aspect ratio doesn't matter.
Maybe I am missing some key points and I need some hints. How should the math for a cubic projection looks like?
The key-point here is: normals shall be in object-space. Please note that gman's answer is more elegant than mine, by using a matrix for the uv
computation. I am using instead the bounding box coordinates, which are already passed to the vertex shader as uniform
for other general purposes.
Moreover, I don't even need to distinguish all the six major axis, I just only need three sides projection, so this can be simplified down. Of course, the texture will be mirrored on the opposite faces.
float sX = u_bbmax.x - u_bbmin.x;
float sY = u_bbmax.y - u_bbmin.y;
float sZ = u_bbmax.z - u_bbmin.z;
/* --- BOX PROJECTION - THREE SIDES --- */
if( (abs(modelNormal.x) > abs(modelNormal.y)) && (abs(modelNormal.x) > abs(modelNormal.z)) ) {
uvCoords = modelPos.yz / vec2(sY, -sZ); // X axis
} else if( (abs(modelNormal.z) > abs(modelNormal.x)) && (abs(modelNormal.z) > abs(modelNormal.y)) ) {
uvCoords = modelPos.xy / vec2(sX, -sY); // Z axis
} else {
uvCoords = modelPos.xz / vec2(sX, -sZ); // Y axis
}
uvCoords += vec2(0.5);
Explanation:
modelPos
coordinates.
Example: the texture can be rotated by 90 degrees by using
modelPos.yx
instead of modelPos.xy
.modelPos
coordinates.
Example: the texture can be mirrored on the Y-axis by using
vec2(sX, sY)
instead of vec2(sX, -sY)
.Result:
EDIT:
It is worth to link here another answer from gman which contain additional information about this topic and also some cool optimization techniques to avoid conditionals inside GLSL shaders: How to implement textureCube using 6 sampler2D.