Why does this RayMarcher doesn't work with 3D Textures?

Development Asset Creation
, , ,
1

Hi,
so I try to use the RayMarcher, which is shown in [here][1].

I use the basic code from the first sample (opacity only).
The only difference is, that I don’t use a 2D Flipbook, but a real Volume Texture which is available sind 4.20 (You can create it from a flipbook, but I also implemented a Factory which imports my raw 3D data, both methods show this bug). When using his flipbook approach, everything is fine. But when I use the Volume Texture I get weird artifacts.

Here is the custom Material Node Code which I use to compute the Emissive Color:

float accumdist = 0;
float3 localCamVec = normalize(mul(Parameters.CameraVector, Primitive.WorldToLocal));
float StepSize = 1 / MaxSteps;

for(int i = 0; i < MaxSteps; i++)
{
float3 satPos = saturate(CurPos);

float curSample = Tex.SampleLevel(TexSampler, satPos, 0).r;

accumdist += curSample * StepSize;
CurPos += -localCamVec * StepSize;
}

return accumdist;

CurPos is BoundingBoxBased_0-1_UVW.

The following artifact is visible:

You can see the stripes which are emitting from the ball.
Another one from the side:

I used Opaque instead of Additive blending so you can see it better. It also goes in the depth.

I have no idea what is wrong since it works with the flipbook. I think maybe it is a bug with Sampling of the new Volume Textures? This also occurs with different Volume Textures.

2

Most likely you are hitting volume limits and as your texture coordinates are saturated, when you are hitting a plane of the bounding box, your raymarch continues with 2 coordinates out of 3 incrementing, resulting in said artifacts.
Either pre-calculate exit point and adjust step number accordingly or check if you are still inside the volume at every step.

3

Thanks, your answer brought me on the right track. I thought this can’t be the problem because I also used his code where he precalculates the Step count. But the article didn’t show that the StepSize has to be precalculated too, otherwise a modified step count doesn’t do anything, because the ray marcher code will always fit the step size to a ray length of 1. The flipbook approach also showed artifacts with other volumes that are more box-like. Those start to repeat when the ray goes beneath the volume.