I want to briefly describe my current reserch focus in this project on this page.
Full implementation of method described in [1] on current Geforce FX class GPU
Some
parts of the paper are brief (for example how to cast multiple
secondary rays) and I need to investigate possible implementation
methods. First implementation will be good platform for tests, tweaks
and future research. I also check if there are no fatal cave-eats for
me.
Investigation of different acceleration structures than uniform grid
I want to focus on:
1. Research of non-uniform acceleration grid strucutre
2. Research of bounding boxes and some kind of LOD system for accelerating ray-triangle intersections
3. Research of some hierarchy acceleration structure
Investigation of use vista buffers
I
want to focus on possibility of use some kinds of vista buffers to
speed up (primary) rays hit. With encoding object ids as colors,
traditional rasterization from eye view, light view or ray view can be
used to speed triangle lookup for intersection.
I want to investigate jezeq's idea of "vista cube-map generation" to
speedup secondary rays.
Investigation of dynamic scene updates
I want investigate possiblities of dynamic changes to data structures stored in GPU. I see main research paths in:
| 1. | Research
of dynamic data structure which can be updated using incremental
texture uploads and simple GPU operation to apply them. | | 2. | Research
of encoding objects geometry in local object space and doing
transformation in pixel shaders. Transformations would be uploaded each
frame. |
Investigation of 'fake' methods to improove image quality or performance
This
chapter is open for now. I want to do specific hacks to fit use in
today's game engine. I'm not aiming physically correct light behaviour.
The performance has higher priority for this project.
Under sampling, texture mapping, partial scene raytracing take place
here.
| 
