I try to split a big image into small patches and use vgg19 to extract features from each patch, and then combine the features for model.
def create_model_tl(inp