aserbremen / powell_ceres

powell_ceres

This repository has the purpose of timing the minimization example https://ceres-solver.googlesource.com/ceres-solver/+/master/examples/powell.cc of the ceres solver (http://ceres-solver.org/).

Three different methods are used to define cost functions and its derivates.

1. Analytic derivation of jacobians using a SizedCostFunction

2. Automatic differentiation defining a single cost functor with 4 residuals

3. Automatic differenttiation defining 4 cost functors (as in the example given by ceres)

Here are the results on my machine in seconds for 10000 runs:

Method	Analytic	Automatic 1 cost function	Automatic 4 cost functions
total time [s]	0.802201	2.1996	2.91902
mean time [s]	0.0000802201	0.00021996	0.000291902

Analytic faster than Automatic 1 cost function by 274.196%
Analytic faster than original Automatic 4 cost functions by 363.876%

This shows that even for small problems optimization time strongly depends on the way cost functions are implemented, and whether analytic jacobians can be supplied or not.