Huai-Ping Lee
Images generated with a global photon map of 70,000 photons, with three different numbers of photon neighbors:
| Nneighbor = 100 | Nneighbor = 150 | Nneighbor = 200 |
|
|
|
Images generated with a global photon map of 70,000 photons and a caustics photon map of 10,000 photons, with three different numbers of photon neighbors:
| Nneighbor = 100 | Nneighbor = 150 | Nneighbor = 200 |
|
|
|
Reference image without global illumination (no supersampling):
The images I generated with photon mapping have the problem of scaling of the power when estimating radiance from photons. I had to divide the radiance by the number of photon neighbors to avoid being too bright (entirely white), but after that the images are a bit dark. Maybe it can be solved with some tweaking. Smooth diffuse surface is very hard to render, and my un-optimized code does not allow me to use too many photons. The bleeding of color around the corners is also obvious.
The code is here. The kd-tree library I used is KDTREE2 by Matthew B. Kennel at Institute for Nonlinear Science, University of California, San Diego.