VolPathIntegrator: track beta and 1 / path MIS weights.
(Rather than T_maj and path PDFs.) This is "just" a matter of dividing the old T_maj by the PDF for sampling the path and then dividing the old path PDFs by that same factor, so the same mathematically in the end. However, it a) better aligns the VolPathIntegrator with the other integrators, which track beta, and b) fixes the issue of T_maj and the PDFs coming to have very large or very small magnitudes and being succeptible to over/underflow. Now, all of the Rescale() stuff is no longer necessary for them...
Showing
想要评论请 注册 或 登录