Add the two final runs of the models' training so so that the loss curves and CER/WER plots on the validation set can be viewed via tensorboard

Add structure for preprocessing of images from dataset such that they all have the same size (as mandated by ResNet)