Position reconstruction from gl_FragCoord and depth value using matrix multiply

[JoshKlint]

I have existing fragment position reconstruction code but it assumes an orthogonal matrix. I’d like to get position reconstruction working with any arbitrary projection matrix. (The variable inverseprojectionmatrix is the inverse of [the projection matrix multiplied by the camera object’s matrix].)

	//Construct gl_FragCoord equivalent
	vec4 fragpos;
	fragpos.xy = gl_FragCoord.xy / BufferSize * 2.0f - 1.0f;// convert to -1, 1 range
	fragpos.z = texelFetch(depthTexture, ivec2(gl_FragCoord.xy), 0).r;// get the depth at this coord
	fragpos.w = 1.0f;

	//Convert fragcoord to gl_Position equivalent
	vec4 glposition = fragpos;
	glposition.z = glposition.z * 2.0f - glposition.w;// opposite of gl_Position.z = (gl_Position.z + gl_Position.w) * 0.5f;

	//Convert gl_Position equivalent to vertex position
	vec3 worldspaceposition = (inverseprojectionmatrix * glposition).xyz;

	//Visualize the result
	outColor.rgb = worldspaceposition / 10.0f;

For clarity, all my vertex shaders do this:

gl_Position = projectionmatrix * vertexworldspaceposition;
gl_Position.z = (gl_Position.z + gl_Position.w) * 0.5f;

Here is what the results should look like using my code I know works:

Here is the result with the code above. It’s close but something is off, and I can’t find any formula for how gl_FragCoord is calculated from gl_Position. Can you tell what is wrong with my code?

[JoshKlint]

It looks like the math all works out if you account for W component of the matrix / vector multiplcation:

//Convert gl_Position equivalent to vertex position
vec4 worldspaceposition = (inversemodelprojectionmatrix * glposition);
worldspaceposition.xyz /= worldspaceposition.w;
worldspaceposition.w = 1.0f;

[krOoze]

You should redivide your w back to 1, as always I think.

Can you walk me throught the good image. Does not quite make sense to me. In your code it looks like you route xyz to rgb. But in that case why is there large uniform green patch (high y?) in the back? Why is there cyan block (high y and z)? Why things close to camera are red (high x)?

PS: Why the texel fetch?

[JoshKlint]

@krOoze

Can you walk me throught the good image. Does not quite make sense to me. In your code it looks like you route xyz to rgb. But in that case why is there large uniform green patch (high y?) in the back? Why is there cyan block (high y and z)? Why things close to camera are red (high x)?

The dragon is at position 0,0,0 and the camera is rotated 90 degrees to the left. Everything looks correct to me.

PS: Why the texel fetch?

Since it’s reading a depth value there is no need for linear filtering. I’m guessing texelFetch could maybe be slightly faster? I don’t think it will be slower at least.

Final code is very simple:

// position.xy = gl_FragCoord.xy / viewportsize
// position.z = depth
vec3 ScreenCoordToWorldPosition(in vec3 position)
{
	vec4 coord = InverseCameraProjectionViewMatrix * vec4(position.xy * 2.0f - 1.0f, position.z * 2.0f - 1.0f, 1.0f);
	return coord.xyz / coord.w;
}

[krOoze]

If the dragon is at 0, then it should be black, no? Anyway seems you solved it and are satisfied, so it doesn’t matter…

I mean, why not just FragCoord.z?