Hands-On Generative Adversarial Networks with Keras

Sampling the generator

We randomly sample the generator to informally evaluate image quality with respect to the constraints described in the text. Inference is straightforward and is described in the following code block:

def infer(data_filepath='data/flowers.hdf5', z_dim=128, out_dir='gan',          n_samples=5):    # we load the saved model    G = load_model(out_dir)    # get text embeddings and text from the validation set    val_data = get_data(data_filepath, 'train')    val_data = next(iterate_minibatches(val_data, n_samples))     emb, txts = val_data[1], val_data[2]    # we sample a z to produce fake images with the text embeddings    z = np.random.uniform(-1, 1, size=(n_samples, z_dim))    G.trainable = False    fake_images = G.predict([z, emb])    # generate n_samples for ...