In calcshrunkencovariance, there isn't really a need to separately compute cov for each of the images and then average. Rather, we can just do some math and figure out how to appropriately scale the cov of the mean-subtracted aggregated residuals. The result is numerically identical. So, this could result in a potentially large speedup.
In calcshrunkencovariance, there isn't really a need to separately compute cov for each of the images and then average. Rather, we can just do some math and figure out how to appropriately scale the cov of the mean-subtracted aggregated residuals. The result is numerically identical. So, this could result in a potentially large speedup.