Cornellbox: Markov Chain Random Walking, Part 2
XianMing Chen, xchen AT cs DOT utah DOT edu
Top
Previous
Next
In the first part, I got some test images of Metropolis algorithm. Those two images are partial solution of constrained path space of
form s1t2.
After I implemented bidirectional path tracing, and therefore any sampling technique of the form s?t? can be used, I come back
to metropolis algorithms. However, it turns out compared to the partial solution, this time there is huge variance if I used all
the sampling strategies of any s?t? forms. So I limited to all those of s1tn(n is any in { 1, 2, 3 ... }), and get some images
as below. The variance is still quite high.
| 25 samples |
 |
| 100 samples |
 |