There is large consent that successful training of deep networks requires many thousand annotated training samples. Olaf Ronneberger, et al., "U-Net: Convolutional Networks for Biomedical Image Segmentation" https://arxiv.org/abs/1505.04597…