I am developing a DNN algorithm and am in need of saving thousands of 2D (maybe 4 channels of 2D) Numpy arrays as labels. 2000 X 2000 array of integers.
I am not sure what is the most efficient way (in terms of speed and storage space) to save these arrays. I have tried: - Numpy save and load, resulting in huge file sizes (hundreds of Mbs for one array) - skimage package's imsave and imload, which is much faster but lossy and changes the numbers when reading back. Generally any package I use for saving image is lossy.
Please advise.