loss_per_training_iterations shows loss per each training loop
Assignment.jpg is segmented using contours method in opencv and divided into marathi/training/ and marathi/testing
siamese network was used as number of images each class was very low i.e one per each letter.
used pytorch as framework