I did not have enough time to read Jensen's original idea. In my implementation, I couple the photon map with the geometry, which
I know is not the way the standard photon mapping does. However, I think my coupling of photon map with the geoemtry
does not put too much limit on geoemtry complexity as it appears. Another problem with my photon mapping is, I cannot get
numberical agreement with (bi)path tracing and metropolis algorithms. The different methods to solve the same integral should give us
the same (statistically) result, and I will try to figure out this problem. All the following images are scaled (with a differert
value from the previous pages) and square-rooted once (which is the same as before).
Comparison of different esitimating photons, all with 1,200,000 tracing photons, 9 samples per pixel
100 photons
500 photons
1000 photons
1500 photons
Comparison of different tracing photons, estimating photons: one thousandth of tracing photons, 9 samples per pixel.
