Two steps as good as three, maybe better?

Everything looks pretty good with two Gauss-Newton steps.

There's a mix of wins, losses, and ambiguous changes here, with many
more wins than losses.  I definitely wouldn't be eager to land the
inverse of this change: slow things down, fit one or two profiles a bit
better, and fit a whole bunch a bit worse.

Change-Id: Icba5d4e0fcda8f668ebd20faab1aec38c6e94106
Commit-Queue: Mike Klein <>
Reviewed-by: Brian Osman <>
34 files changed