Discussion points raised by Matt A:

1) Maloney et al. imply that illumination estimation is a critical step. What other mechanisms could explain their data?

2) What are some of the reasons to be skeptical about using Bayes rule to simulate vision?