Please sign in to comment.
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
reverted to L2 regularization for splined models
The difference-from-mean regularization wasn't good as a "regularization". Any place would be a perfect fit as far as the regularization is concerned. So the optima found by the solver would end up with giant focal lengths, compensated for by crazy intrinsics
- Loading branch information
Oops, something went wrong.