Cropped cells, depending from the whole image they are sampled from, have different intensities. We need to normalize the cropped cells image-wise.

Todo:

• Image-wise, channel-wise normalization for each cropped cell (image number is the first number of the file name)
• Reading images into 60 x 60 x 3 numpy matrices ( if we use a convnet encoder and deconv decoder, or should we use a fully connected vanilla VAE? )