Image2Image Neural Network (Deep Learning)
7 views (last 30 days)
Hello I would like to implement a image2image NN. During training I would like to feed my network 2 images (Before & After) so that it can learn the manipulation (i.e. Denoising, filtering, correction, etc..).
I have built my network and tested it on a small dataset (100 images, each 64x64). Currently when training my network I must load in (ALL) the data into memory.
Does the trainNetwork fcn have the ability to handle a tbl input where the first column is a list of filepaths to my before files and the second column is a list of filepaths to my after files?
if not - what are my other options in terms of passing though a list of filenames for before/after images into my network?
These images must be TIFF to ensure 32bit files
Johanna Pingel on 6 Feb 2019
Try following this example:
The idea is to use mini-batch datastore to create the before and after images, which should handle the memory management for you.
Let me know if this doesn't solve your problem.