I am working on a deep learning problem which is cancer aid detection, in which predictors and response (i.e. features and labels) are images with size 250×250. I have hardware limitations and I run out of memory, after which I decided to convert images to Matlab formatted files (“.mat”), it made some good enhancements. However, memory issue still exists. I have explored some resources in which they highly recommend using numpy files (“.npy”).
As it would be a bit costly to do that change (transforming images to numpy files) I needed to make sure first that it would really make a difference.
I am not asking for memory enhancements algorithms (e.g. batching). Just the difference between “.mat” and “.npy”.