Question5-3.txt

Effect 1: Photon mapping is a biased estimation. As the result, the final image regardless
of number of iterations does not converge to the desired image. Although one can claim that
rendering a plausible image is possible by infinitely many photons in the scene. The variance
of photon mapping is small so the error reduces faster. Path tracing converges to the desired
image but it takes a very long time (in this case we still do not have a plausible image after
100 or 500 iterations while we get a plausible image by photo mapping in only 10 iterations)

Effect 2: The noise in the cornell_box_water scene for path tracing in the water area is much
than corresponding area in the image rendered by photon mapping. The reason is, in path tracing
when the ray hits the floor, then the next event estimator cannot help in the convergence because
there is an object on top of it (surface of the water) which causes an occlusion. Thus, another
bounce must be traced by the BRDF density function which hardly has a chance to end up on the light
source.
But in photon mapping, this issue does not exist since there are photons on the floor already
(refracted by water) before we do the ray tracing.