dcgan.py

from keras.datasets import mnist
from keras.layers import Dense, Reshape, Conv2D, Flatten, Input, UpSampling2D
from keras import Sequential, Model
from keras.optimizers import Adam
import numpy as np
import matplotlib.pyplot as plt
import keras.backend as K
K.clear_session()

input_size = 100  # 100 random sayi
optimizer = Adam(lr=0.0002)

###
generator = Sequential()
generator.add(Dense(784, activation='linear', input_dim=input_size))
generator.add(Reshape([7, 7, 16]))
generator.add(UpSampling2D())
generator.add(Conv2D(32, kernel_size=5, strides=1, padding='same'))
generator.add(UpSampling2D())
generator.add(Conv2D(1, kernel_size=5, strides=1,
                     padding='same', activation='tanh'))
generator.summary()

discriminator = Sequential()
discriminator.add(Conv2D(64, kernel_size=5, strides=2,
                         padding='same', input_shape=(28, 28, 1)))
discriminator.add(Conv2D(32, kernel_size=5, strides=2, padding='same'))
discriminator.add(Flatten())
discriminator.add(Dense(32, activation='relu'))
discriminator.add(Dense(1, activation='sigmoid'))
discriminator.compile(loss="binary_crossentropy", optimizer=optimizer)
discriminator.summary()
###

discriminator.trainable = False
x = Input((input_size,))
out_generator = generator(x)
out_discriminator = discriminator(out_generator)
gan = Model(inputs=(x,), outputs=(out_discriminator))
gan.compile(loss="binary_crossentropy", optimizer=optimizer)
gan.summary()

def generateRandomData(sizey, sizex):
    return np.random.normal(0, 1, (sizey, sizex))

def showResults(gen):
    noise = generateRandomData(32, input_size)
    images = gen.predict(noise)
    print images.shape
    print images[0].dtype
    plt.figure(figsize=(4, 8))
    for i in range(32):
        plt.subplot(4, 8, i + 1)
        im = np.reshape(images[i], (1, -1))
        im = (np.reshape(im, [28, 28]) + 1) * 255
        im = np.clip(im, 0, 255)
        im = np.uint8(im)
        plt.imshow(im, cmap='gray')
    plt.show()

(xtrain, _), (xtest, _) = mnist.load_data()
###
xtrain = np.reshape(xtrain, [-1, 28, 28, 1])
xtest = np.reshape(xtest, [-1, 28, 28, 1])
###
xtrain = (xtrain.astype(np.float32) - 127.5) / 127.5
# plt.imshow(np.reshape(xtrain[0], [28, 28]), cmap="gray")
# plt.show()


epochs = 20
batch_size = 128
eval_size = 32

print xtrain.shape, xtest.shape
for e in range(epochs):
    for i in range(xtrain.shape[0] / batch_size):
        # gercek veriyi al
        xreal = xtrain[(i) * batch_size:(i + 1) * batch_size]
        # fake veriyi olustur
        noise = generateRandomData(batch_size, input_size)
        # fake ciktiyi al
        xfake = generator.predict_on_batch(noise)
        # gercek ile egit
        discriminator.trainable = True
        discriminator.train_on_batch(xreal, np.array([0.9] * batch_size))
        # fake ile egit
        discriminator.train_on_batch(xfake, np.array([0.] * batch_size))
        # gan i egit
        discriminator.trainable = False
        gan.train_on_batch(noise, np.array([1.] * batch_size))

    # ekrana yazdir
    if (e + 1) % 4 == 0:
        showResults(generator)