How Simpleitk calculate the gain factor and the approximation gradient for stochastic gradient descent?

Hi, I as looking for how to use stochastic gradient descent and I found that it can be done by setting MetricSamplingStrategyType as RANDOM and that the adaptative gain factor can be done by setting EstimateLearningRateType as EachIteration.

My problem is that I looked in the documentation and in the simpleitk papers and didn’t find the methods for calculating the gradient and the gain factor. So, does anyone know where I can find this?


Hello @Felippe_Trigueiro_An,

As SimpleITK is a wrapper around ITK, the actual implementation is in the ITK code. The theory behind the ITK registration implementation was described in this paper.

1 Like