Closed ircwaves closed 6 years ago
I think I believed that doing the arithmetic T * Ld * scaling_factor
outside of numpy would be faster because I believed ((rad_rev_img[c] - Lu) / (T * Ld * scaling_factor)
) would result in many repetitions of T * Ld * scaling_factor
. I don't remember if I observed it actually going faster in an A/B test, though.
Tom -- can you clarify what you mean in this comment: https://github.com/Applied-GeoSolutions/gips/blob/a2bd9a60a9d384e73404b1c21781e24facd54bfd/gips/data/sentinel2/sentinel2.py#L957