It seems that cm_predict currently assumes that all the predictions fit in RAM. This might not always be the case, and I think it would be reasonable to predict in batches, store the results batch by batch, and at the end mosaick all the sub-tiles together.
It seems that cm_predict currently assumes that all the predictions fit in RAM. This might not always be the case, and I think it would be reasonable to predict in batches, store the results batch by batch, and at the end mosaick all the sub-tiles together.
What do you think?