Thanks Matt.
We are keen on using ICA-AROMA as our own internal benchmarking on non-multiband data has shown that it does quite a good job of denoising, without the need to train a new classifier.
My main question is why the results of ICA-AROMA should vary so much (i.e., correlation of only ~.6) between different runs of the algorithm on the exact same data, and whether there is a specific reason why this should occur for multiband data.
Kristina
|