At this point, we usually implement the train method, but because CycleGAN has more components, we'll show you how to build the entire model. First, we instantiate the data_loader object, where you can specify the name of the training set (feel free to experiment with the different datasets). All the images will be resized to img_res=(IMG_SIZE, IMG_SIZE) for the network input, where IMG_SIZE = 256 (you can also try 128 to speed up the training process):
# Input shapeimg_shape = (IMG_SIZE, IMG_SIZE, 3)# Configure data loaderdata_loader = DataLoader(dataset_name='facades', img_res=(IMG_SIZE, IMG_SIZE))
Then, we'll define the optimizer and the loss weights:
lambda_cycle = 10.0 # Cycle-consistency losslambda_id = 0.1 ...
