implement volume rendering with object order empty space skipping

Question

I learn some volume rendering techniques from the Advanced IlluminationTechniques forGPU-Based Volume Raycasting of scenegraph 2009 course note. I am interested in the object order empty space skipping, which rasterizing the bounding box into lots of small blocks, like the image bellow:

I want to know how to implement this? by octree ? or any other techniques? I have implemented a simple ray casting volume rendering program without object order empty space skipping. code snippet or pseudocode would help a lot.

Answer 1

The basic idea is to generate a bounding geometry as a triangle mesh, and then render that mesh to an off-screen buffer in order to compute the start and end positions that are used for the ray marching in the second pass.

I've implemented this in C++ using OpenGL 4 and blogged about it, including source code. Here's the general outline:

• Initialization: Given a volume texture with iso values, generate a bounding geometry on a coarse grid. For example, the volume could be 256x256x200, and the grid cells could span 8³ voxels, so the grid dimensions would be 32x32x25.
  • First generate a volume texture gridTex, where each voxel contains the maximum iso value.
  • Given a threshold iso value, generate the bounding geometry. A grid voxel is considered active if it contains anything visible. For iso-surfaces that means the value is above the threshold. For each pair of neighboring voxels in gridTex, generate a quad if one is active and the other is inactive. Do this every time the iso value changes.
• First pass: Render the bounding geometry to an offscreen buffer, storing the minimum and maximum depth values for each fragment.
• Second pass: Render the front faces again to the main buffer.
  • Discard if not at the minimum depth value. Compute the start and end positions for the ray from the minimum and maximum depth values.
  • March the ray as usual.