SDFRaymarch - Read the depth value

[RenaudRohlinger]

I'm looking for a way to get the correct depth value of a raymarching result. I will open an issue here as I think it could benefit other developers too.

I made an example with a post process splitting the screen into 4 sections:
1 - scene albedo
2 - raymarching albedo
3 - scene depth
4 - raymarching depth (using custom gl_FragDepth = length(point.xyz))

https://codesandbox.io/s/romantic-sunset-uqhc13?file=/src/RayMarchSDFMayerial.js

[gkjohnson]

Hmm - I'm not able to see the raymarch depth buffer on my machine. Is there something I need to enable?

[RenaudRohlinger]

Oh, my bad indeed!

It seems that some WebGL Implementations do not like the texelFetch syntax (Metal for instance). I changed it with a basic texture()

readDepthFrag(texelFetch(sceneDepth, fCoord, 0).x => texture(sceneDepth, vUv).x

It should be fixed @gkjohnson 👍

https://codesandbox.io/s/romantic-sunset-uqhc13?file=/src/RayMarchSDFMayerial.js

[CodyJasonBennett]

Looks like I get a correct result with viewZToPerspectiveDepth, initializing fragCoordZ = -1. Taking note of the math linked from https://github.com/mrdoob/three.js/blob/bb23058464420d8fd510337ff400182ba662e8dd/src/renderers/shaders/ShaderChunk/packing.glsl.js#L54.

float cameraNear = 0.1;
float cameraFar = 5.0;

gl_FragDepth = viewZToPerspectiveDepth(-(fragCoordZ / gl_FragCoord.w), cameraNear, cameraFar);

[gkjohnson]

Thanks @CodyJasonBennett this is great!

A few changes to make still, I think. You shouldn't need to do anything with fragCoord.w since the raymarching is happening in view space (no projection matrix applied) and the w coordinate should be 1.0 anyway since it's just a full screen quad being rendered in that pass.

And I believe the fragCoordZ should be set to point.z rather than length( point.xyz ) (my bad 😁) since it just needs to test z distance for depth:

float fragCoordZ = - 1.0;

// ...

if ( intersectsSurface ) {

  fragCoordZ = - point.z;
  // ...

}

// ...

gl_FragDepth = viewZToPerspectiveDepth( - fragCoordZ, cameraNear, cameraFar );

The depth visually matches a lot more precisely after these changes though more testing should still be done to make sure the type of shifting you originally showed me is addressed @RenaudRohlinger.

For performance reasons in the raymarch loop, though, I would transform the scene depth from the existing depth buffer beforehand to check in the for loop.

[CodyJasonBennett]

Awesome, I had to set fragCoordZ = -point.z, but it works nicely.

[gkjohnson]

Awesome, I had to set fragCoordZ = -point.z, but it works nicely.

Oops - you're right. I'd inadvertently been taking the absolute value of point.z. I've updated my notes above to specify your change. Thanks!

Tough I guess we could just remove the negation in all the places, instead 😅

[RenaudRohlinger]

Awesome! Thank you so much for this guys, it's working perfectly 😊

[gkjohnson]

Awesome! Thank you so much for this guys, it's working perfectly 😊

Twitter pics please! 🙏