Fast bare-bones implementation of convolutional layers, residual blocks, Adam optimizer, backpropagation and custom accuracy and loss functions (F1 score per pixel and binary-crossentropy)
