Reconstructing world position from linear depth

1.4k views Asked by At

i have issues reconstructing world positions from previously stored linear depth in glsl. I read lots of info online, but can't find my problem... So this is what I got:

VS (storing depth to 32F):
float linDepth(float z) {
    return (-z-zNear)/(zFar-zNear);
}
void main() {
    vec4 position = uViewMatrix * uModelMatrix * vec4(aPosition, 1);
    depth = linDepth(position.z); //store linear view-depth
}

FS (reconstuction):
void main() {
    vec3 vUV = vec2(0..1, 0..1); (from screen aligned quad)
    vec3 ndc = vec3(vUV*2-1, linearViewDepth*2-1);
    vec4 v0 = inverse(uProjectionMatrix)*vec4(ndc, 1);
    vec3 reconViewPos = v0.xyz/v0.w;
    vec3 reconWorldPos = inverse(uViewMatrix) * v0;
}

...and the results are completely off. Though I sense a problem by using unchanged linear view depth as ndc z. In the end I wish to apply the ray-interpolation approach:

VS (reconstruction, aligned screenquad):
out vec3 vViewRay;
void main() {
    gl_Position = aPosition;
    vec4 v = vec4(aPosition.x, aPosition.y, 1, 1); //ndc (at the back of cube)
    v = inverse(uProjectionMatrix) * v;
    v /= v.w; //view coordinate
    v /= v.z; //normalize by z for scaling
    vViewRay = v.xyz;
}

FS(reconstruction):
in vec3 vViewRay;
float delinDepth(float z) {
    return -(z*(zFar-zNear)+zNear);
}
void main() {
    vec3 reconViewPos = vViewRay * delinDepth(linearViewDepth);
}
1

There are 1 answers

0
Teris On BEST ANSWER

In the end the second reconstruction code did work, I guess something was off somewhere else...