diff options
| -rwxr-xr-x | cdcgan.py | 233 |
1 files changed, 233 insertions, 0 deletions
diff --git a/cdcgan.py b/cdcgan.py new file mode 100755 index 0000000..aba3669 --- /dev/null +++ b/cdcgan.py @@ -0,0 +1,233 @@ +from __future__ import print_function, division +import tensorflow.keras as keras +import tensorflow as tf +from keras.datasets import mnist + +import keras.layers as layers +from keras.layers import Input, Dense, Reshape, Flatten, Dropout, multiply +from keras.layers import BatchNormalization, Activation, Embedding, ZeroPadding2D +from keras.layers.advanced_activations import LeakyReLU +from keras.layers.convolutional import UpSampling2D, Conv2D +from keras.models import Sequential, Model +from keras.optimizers import Adam +import matplotlib.pyplot as plt +from IPython.display import clear_output +from tqdm import tqdm + +import numpy as np + +class CDCGAN(): + def __init__(self): + # Input shape + self.img_rows = 28 + self.img_cols = 28 + self.channels = 1 + self.img_shape = (self.img_rows, self.img_cols, self.channels) + self.num_classes = 10 + self.latent_dim = 100 + + optimizer = Adam(0.0002, 0.5) + + # Build and compile the discriminator + self.discriminator = self.build_discriminator() + + self.discriminator.compile(loss=['binary_crossentropy'], + optimizer=optimizer, + metrics=['accuracy']) + + # Build the generator + self.generator = self.build_generator() + + # The generator takes noise and the target label as input + # and generates the corresponding digit of that label + noise = Input(shape=(self.latent_dim,)) + label = Input(shape=(1,)) + img = self.generator([noise, label]) + + # For the combined model we will only train the generator + self.discriminator.trainable = False + + # The discriminator takes generated image as input and determines validity + # and the label of that image + valid = self.discriminator([img, label]) + + # The combined model (stacked generator and discriminator) + # Trains generator to fool discriminator + self.combined = Model([noise, label], valid) + self.combined.compile(loss=['binary_crossentropy'], + optimizer=optimizer) + + def build_generator(self): + # Prepare noise input + input_z = layers.Input((100,)) + dense_z_1 = layers.Dense(1024)(input_z) + act_z_1 = layers.Activation("tanh")(dense_z_1) + dense_z_2 = layers.Dense(128 * 7 * 7)(act_z_1) + bn_z_1 = layers.BatchNormalization()(dense_z_2) + reshape_z = layers.Reshape((7, 7, 128), input_shape=(128 * 7 * 7,))(bn_z_1) + + # Prepare Conditional (label) input + input_c = layers.Input((1,)) + dense_c_1 = layers.Dense(1024)(input_c) + act_c_1 = layers.Activation("tanh")(dense_c_1) + dense_c_2 = layers.Dense(128 * 7 * 7)(act_c_1) + bn_c_1 = layers.BatchNormalization()(dense_c_2) + reshape_c = layers.Reshape((7, 7, 128), input_shape=(128 * 7 * 7,))(bn_c_1) + + # Combine input source + concat_z_c = layers.Concatenate()([reshape_z, reshape_c]) + + # Image generation with the concatenated inputs + up_1 = layers.UpSampling2D(size=(2, 2))(concat_z_c) + conv_1 = layers.Conv2D(64, (5, 5), padding='same')(up_1) + act_1 = layers.Activation("tanh")(conv_1) + up_2 = layers.UpSampling2D(size=(2, 2))(act_1) + conv_2 = layers.Conv2D(1, (5, 5), padding='same')(up_2) + act_2 = layers.Activation("tanh")(conv_2) + model = Model(inputs=[input_z, input_c], outputs=act_2) + return model + + + def build_discriminator(self): + input_gen_image = layers.Input((28, 28, 1)) + conv_1_image = layers.Conv2D(64, (5, 5), padding='same')(input_gen_image) + act_1_image = layers.Activation("tanh")(conv_1_image) + pool_1_image = layers.MaxPooling2D(pool_size=(2, 2))(act_1_image) + conv_2_image = layers.Conv2D(128, (5, 5))(pool_1_image) + act_2_image = layers.Activation("tanh")(conv_2_image) + pool_2_image = layers.MaxPooling2D(pool_size=(2, 2))(act_2_image) + + input_c = layers.Input((1,)) + dense_1_c = layers.Dense(1024)(input_c) + act_1_c = layers.Activation("tanh")(dense_1_c) + dense_2_c = layers.Dense(5 * 5 * 128)(act_1_c) + bn_c = layers.BatchNormalization()(dense_2_c) + reshaped_c = layers.Reshape((5, 5, 128))(bn_c) + + concat = layers.Concatenate()([pool_2_image, reshaped_c]) + + flat = layers.Flatten()(concat) + dense_1 = layers.Dense(1024)(flat) + act_1 = layers.Activation("tanh")(dense_1) + dense_2 = layers.Dense(1)(act_1) + act_2 = layers.Activation('sigmoid')(dense_2) + model = Model(inputs=[input_gen_image, input_c], outputs=act_2) + return model + + def train(self, epochs, batch_size=128, sample_interval=50, graph=False, smooth_real=1, smooth_fake=0): + + # Load the dataset + (X_train, y_train), (_, _) = mnist.load_data() + + # Configure input + X_train = (X_train.astype(np.float32) - 127.5) / 127.5 + X_train = np.expand_dims(X_train, axis=3) + y_train = y_train.reshape(-1, 1) + + # Adversarial ground truths + valid = np.ones((batch_size, 1)) + fake = np.zeros((batch_size, 1)) + + xaxis = np.arange(epochs) + loss = np.zeros((2,epochs)) + for epoch in tqdm(range(epochs)): + + # --------------------- + # Train Discriminator + # --------------------- + + # Select a random half batch of images + idx = np.random.randint(0, X_train.shape[0], batch_size) + imgs, labels = X_train[idx], y_train[idx] + + # Sample noise as generator input + noise = np.random.normal(0, 1, (batch_size, 100)) + + # Generate a half batch of new images + gen_imgs = self.generator.predict([noise, labels]) + + # Train the discriminator + d_loss_real = self.discriminator.train_on_batch([imgs, labels], valid*smooth_real) + d_loss_fake = self.discriminator.train_on_batch([gen_imgs, labels], valid*smooth_fake) + d_loss = 0.5 * np.add(d_loss_real, d_loss_fake) + + # --------------------- + # Train Generator + # --------------------- + + # Condition on labels + sampled_labels = np.random.randint(0, 10, batch_size).reshape(-1, 1) + # Train the generator + g_loss = self.combined.train_on_batch([noise, sampled_labels], valid) + + # Plot the progress + #print ("%d [D loss: %f, acc.: %.2f%%] [G loss: %f]" % (epoch, d_loss[0], 100*d_loss[1], g_loss)) + loss[0][epoch] = d_loss[0] + loss[1][epoch] = g_loss + + # If at save interval => save generated image samples + if epoch % sample_interval == 0: + self.sample_images(epoch) + + if graph: + plt.plot(xaxis,loss[0]) + plt.plot(xaxis,loss[1]) + plt.legend(('Discriminator', 'Generator'), loc='best') + plt.xlabel('Epoch') + plt.ylabel('Binary Crossentropy Loss') + + def sample_images(self, epoch): + r, c = 2, 5 + noise = np.random.normal(0, 1, (r * c, 100)) + sampled_labels = np.arange(0, 10).reshape(-1, 1) + + #using dummy_labels would just print zeros to help identify image quality + #dummy_labels = np.zeros(32).reshape(-1, 1) + + gen_imgs = self.generator.predict([noise, sampled_labels]) + + # Rescale images 0 - 1 + gen_imgs = 0.5 * gen_imgs + 0.5 + + fig, axs = plt.subplots(r, c) + cnt = 0 + for i in range(r): + for j in range(c): + axs[i,j].imshow(gen_imgs[cnt,:,:,0], cmap='gray') + axs[i,j].set_title("Digit: %d" % sampled_labels[cnt]) + axs[i,j].axis('off') + cnt += 1 + fig.savefig("images/%d.png" % epoch) + plt.close() + + def generate_data(self): + noise_train = np.random.normal(0, 1, (55000, 100)) + noise_test = np.random.normal(0, 1, (10000, 100)) + noise_val = np.random.normal(0, 1, (5000, 100)) + + labels_train = np.zeros(55000).reshape(-1, 1) + labels_test = np.zeros(10000).reshape(-1, 1) + labels_val = np.zeros(5000).reshape(-1, 1) + + for i in range(10): + labels_train[i*5500:] = i + labels_test[i*1000:] = i + labels_val[i*500:] = i + + train_data = self.generator.predict([noise_train, labels_train]) + test_data = self.generator.predict([noise_test, labels_test]) + val_data = self.generator.predict([noise_val, labels_val]) + + labels_train = keras.utils.to_categorical(labels_train, 10) + labels_test = keras.utils.to_categorical(labels_test, 10) + labels_val = keras.utils.to_categorical(labels_val, 10) + + return train_data, test_data, val_data, labels_train, labels_test, labels_val + +''' +if __name__ == '__main__': + cgan = CDCGAN() + cgan.train(epochs=70, batch_size=32, sample_interval=200) + train, test, tr_labels, te_labels = cgan.generate_data() + print(train.shape, test.shape) +''' |