- Rotating highlights?
- Can be caused by calculating normals using ray-surface intersection points in world coordinate system instead of canonical coordinate system
- Problem
- Basic Monte Carlo integration is easy:
I = 0 for 1:n x = rand w/ distribution favoring large values of f(x) I = I + f(x)/p(x) I = I / n - But distribution for random numbers does have to favor values
where f(x) is significant (aka “importance sampling”),
otherwise you’ll waste lots of time adding very small
numbers to the estimate
- Basic Monte Carlo integration is easy:
- Inversion
- Use inverse of CDF to map uniformly distributed random numbers
to ones distributed per CDF
u = rand x = CDF-1(u) - Problem is that CDFs are usually hard to invert
- Use inverse of CDF to map uniformly distributed random numbers
to ones distributed per CDF
- Rejection sampling