def main():
## Instantiating generator class and loading model
generator = networks.GeneratorNet()
generator.load_state_dict(torch.load(MODEL_FOLDER))
## Running the sample(s) generation loop
for sample in range(num_test_samples):
## Generating noise to generate images out of them
random_noise = support.noise(num_test_samples)
generated_images = support.vectors_to_images(generator(random_noise))
generated_images = generated_images.detach().numpy()
# Save image
scipy.misc.imsave('generated_images/gen_image_{}.jpg'.format(sample),
generated_images[0][0])
preprocess_image_train,
num_parallel_calls=AUTOTUNE).cache().shuffle(BUFFER_SIZE).batch(
args.batch_size))
train_y = (train_y.map(
preprocess_image_train,
num_parallel_calls=AUTOTUNE).cache().shuffle(BUFFER_SIZE).batch(
args.batch_size))
test_x = test_x.map(preprocess_image_test).batch(NUM_SAMPLES)
test_y = test_y.map(preprocess_image_test).batch(NUM_SAMPLES)
ds = tf.data.Dataset.zip((train_x, train_y))
sample_x = next(iter(test_x))
sample_y = next(iter(test_y))
generator_g = networks.GeneratorNet()
generator_f = networks.GeneratorNet()
discriminator_x = networks.DiscriminatorNet()
discriminator_y = networks.DiscriminatorNet()
generator_g_optimizer = tf.keras.optimizers.Adam(args.lr_generator,
beta_1=0.5)
generator_f_optimizer = tf.keras.optimizers.Adam(args.lr_generator,
beta_1=0.5)
discriminator_x_optimizer = tf.keras.optimizers.Adam(args.lr_discriminator,
beta_1=0.5)
discriminator_y_optimizer = tf.keras.optimizers.Adam(args.lr_discriminator,
beta_1=0.5)
[2] Improved GAN (one-sided label smoothing, minibatch discrimination layer)
[3] Unrolled GAN (not yet implemented)
[4] DCGAN
""")
arch = architectures[int(input())-1]
print("""Choose data:
[1] Cats
[2] Cars (not yet implemented)
""")
dataname = data[int(input())-1]
if (arch == "VGAN"):
G = networks.GeneratorNet()
elif (arch == "GAN_improve"):
G = networks_improve.GeneratorNet()
elif(arch == "DCGAN"):
pass; # TODO: implement me later
for _,_,fnames in os.walk("./data/models/{}".format(arch)):
model_list = fnames #sorted(fnames, reverse = True)
model_list = filter(lambda x: "G" in x, model_list)
max_G = (sorted(model_list, key = lambda x: int(x[x.rindex("_")+1:]))[-1])
max_epoch = max_G[max_G.rindex("_")+1:]
print("Choose epoch: [0 - " + str(max_epoch) + "]")