Graphics Reference
In-Depth Information
The algorithm called Hi-Z Screen-Space Cone Tracing proposed in this chap-
ter can reflect an entire scene with quick convergence and performs orders of
magnitude faster than the linear constant step-based ray-marching algorithm.
4.4 Algorithm
The proposed algorithm can be divided into five distinct steps:
1. Hi-Z pass,
2. pre-integration pass,
3. ray-tracing pass,
4. pre-convolution pass,
5. cone-tracing pass.
We will go through each step by step now.
4.4.1 Hi-Z Pass
The
Hierarchical-Z buffer
, also known as the Hi-Z buffer, is constructed by taking
the minimum or maximum of the four neighboring values in the original Z-buffer
and storing it in a smaller buffer at half the size. In our case for this chapter, we
will go with the minimum version.
The Z-buffer holds the depth values of the 3D scene in a buffer such as a
texture/image. The figure below represents the minimum value version of how a
Hi-Z construction works:
Before
After
The result is a coarse representation of the original buffer. We do this consecu-
tively on the resulting buffers until we arrive at a buffer with the size of 1, where
we no longer can go smaller. We store the computed values in the mip-channels
of a texture. This is represented in Figure 4.6.
The result is what we call a Hi-Z buffer, because it represents the Z values
(also known as scene depth values) in a hierarchical fashion.
This buffer is the heart of the algorithm. It's essentially a screen/image aligned
quad tree that allows us to accelerate the ray-tracing algorithm by noticing and
