Project 3: Panoramas
Low res photo:
When we first started writing ComputeHomography, we attempted to choose the minimum nonzero eigenvalue to determine the row of V to use for our homography, but it turned out that this was not what we wanted.
If we actually have an eigenvalue of zero, this is much
better than a nonzero eigenvalue because it means we managed to fit the points
exactly. It was also unnecessarily computationally intensive since the Eigen
library
already provided the columns in sorted order.