1 - Photon map
A map of the photons stored in the scene
2 - Nearest neighbour search by kd-tree
Suppose we have an 100x100 image (10000 pixels) and for each pixel's primary ray intersection with the scene, we need to find its 100 nearest photons. Let's vary the number of photons used in the photon map, and see how well the naive algorithm and the kd-tree algorithm perform in finding the nearest neighbours:
With 1000 photon casted:
naive: 1.67 seconds
kd-tree: 0.39 seconds
With 10000 photon casted:
naive: 17.54 seconds
kd-tree: 1.60 seconds
With 100000 photon casted:
naive: 167.84 seconds
kd-tree: 20.09 seconds
3 - Filtering photons
Each pixel is coloured according to the intensity of photons around it. Notice the diffuse
reflection off the red and green walls onto the white wall.
4-5 - Reflection, transparency and refraction
Spheres with increasing reflectance (0.1 to 0.5 to 0.9)
Spheres with increasing transmittance/transparency (0.1 to 0.5 to 0.9)
Transparent spheres with increasing indices of refraction (1.2 to 2 to 10)
Mirror and glass spheres in the Cornell Box
6 - Additional primitives
Cylinder, cone and torus
7 - Constructive Solid Geometry
Unions (head and torso), differences (ears) and intersections (hands) of spheres
8 - Texture Mapping
Aliens got tired of abducting macho cows
9 - Anti-aliasing by stochastic sampling
Before anti-aliasing
Pixels requiring anti-aliasing are highlighted in cyan
After anti-aliasing
10 - Final Scenes
Abstract NPR diamond
Spinning "diamond"
Actual diamond!!! Now let's see if it exhibits the hearts and arrows phenomenon
In motion
Top view (partial arrows)
Spin
Bottom view (no hearts)
Whee
11 - Bugs I encountered
texture gets blended in photon mapping
some problem with local/global variables
mirror reflection bug
incorrect use of nearest neighbours
this is probably not right, but I have no intuition for glass tori
Supposed to be a metal sphere I think
Diamond with baguette cut. This is not really a bug but
I ran out of time to make it look nice.
ahh huge bug!