Exposing the DCGAN model for Cifar 10 with keras

Overview

――Since I don't have time, I will omit the mechanism of GAN for the time being. --Write generator and discriminator settings --Miso uses LeakyReLu ** as the activation function for both generator and discriminator. --The source code of the entire learning process will be uploaded to github (sorry for preparation)

Model parameters

Generator

generator

def _build_generator(self) -> Model:
    start_pix_x = 4
    start_pix_y = 4
    kernel_ini = RandomNormal(mean=0.0, stddev=0.02)

    inputs = Input(shape=self.noise_shape)
    x = Dense(
        units=256*start_pix_x*start_pix_y,
        kernel_initializer=kernel_ini,
        bias_initializer='zeros')(inputs)
    x = LeakyReLU(alpha=0.2)(x)
    x = Reshape((start_pix_x, start_pix_y, 256))(x)
    x = Conv2DTranspose(
        filters=128,
        kernel_size=4,
        strides=2,
        padding='same',
        kernel_initializer=kernel_ini,
        bias_initializer='zeros')(x)
    x = LeakyReLU(alpha=0.2)(x)
    # x = BatchNormalization(axis=3)(x)
    x = Conv2DTranspose(
        filters=128,
        kernel_size=4,
        strides=2,
        padding='same',
        kernel_initializer=kernel_ini,
        bias_initializer='zeros')(x)
    x = LeakyReLU(alpha=0.2)(x)
    # x = BatchNormalization(axis=3)(x)
    x = Conv2DTranspose(
        filters=128,
        kernel_size=4,
        strides=2,
        padding='same',
        kernel_initializer=kernel_ini,
        bias_initializer='zeros')(x)
    x = LeakyReLU(alpha=0.2)(x)
    x = Conv2D(
        filters=3,
        kernel_size=3,
        padding='same',
        kernel_initializer=kernel_ini,
        bias_initializer='zeros')(x)

    y = Activation('tanh')(x)

    model = Model(inputs, y)
    if self.verbose:
        model.summary()

    return model

Discriminator

discriminator

def _build_discriminator(self) -> Model:
    kernel_ini = RandomNormal(mean=0.0, stddev=0.02)
    inputs = Input(shape=self.shape)
    x = GaussianNoise(stddev=0.05)(inputs)  # prevent d from overfitting.
    x = Conv2D(
        filters=64,
        kernel_size=3,
        padding='SAME',
        kernel_initializer=kernel_ini,
        bias_initializer='zeros')(x)
    x = LeakyReLU(alpha=0.2)(x)
    # x = Dropout(0.5)(x)
    x = Conv2D(
        filters=128,
        kernel_size=3,
        strides=2,
        padding='SAME',
        kernel_initializer=kernel_ini,
        bias_initializer='zeros')(x)
    x = LeakyReLU(alpha=0.2)(x)
    # x = Dropout(0.5)(x)
    # x = BatchNormalization(axis=3)(x)
    x = Conv2D(
        filters=128,
        kernel_size=3,
        strides=2,
        padding='SAME',
        kernel_initializer=kernel_ini,
        bias_initializer='zeros')(x)
    x = LeakyReLU(alpha=0.2)(x)
    # x = Dropout(0.5)(x)
    # x = BatchNormalization(axis=3)(x)
    x = Conv2D(
        filters=256,
        kernel_size=3,
        strides=2,
        padding='SAME',
        kernel_initializer=kernel_ini,
        bias_initializer='zeros')(x)
    x = LeakyReLU(alpha=0.2)(x)

    x = Flatten()(x)
    features = Dropout(0.4)(x)

    validity = Dense(1, activation='sigmoid')(features)

    model4d = Model(inputs, validity)
    model4g = Model(inputs, validity)
    if self.verbose:
        model4d.summary()

    return model4d, model4g

Output result

The output line corresponds to the class. DCGAN only generates an image, but the generated image is input to the learning model constructed from the original image, labeled according to the predicted label, and the image is output for each prediction class. By putting LeakyReLU in the generator, I feel that the object of the object can be generated more firmly.

Conclusion

I did it in a rush, so I'll write it in detail at a later date.