Vanilla Gan Tutorial!

%%javascript
IPython.OutputArea.auto_scroll_threshold = 9999;

from tensorflow.keras.datasets import mnist

(x_train, y_train), (x_test, y_test) = mnist.load_data()

generator에서 tanh를 activation으로 활용합니다.

tanh를 활용하면 output이 -1 ~ 1 사이로 나오기 때문에 Normalize를 해줄 때 127.5로 나눈 뒤 1을 빼줍니다.

x_train = x_train / 127.5 - 1
x_test = x_test / 127.5 - 1

min 값과 max 값이 -1 ~ 1사이의 범위를 가져야 합니다.

x_train.min(), x_train.max()

(-1.0, 1.0)

x_train 값은 현재 28 * 28로 되어 있습니다. 이를 Flatten하게 펴줍니다.

x_train = x_train.reshape(-1, 784)
x_train.shape

(60000, 784)

필요한 모듈을 import 합니다.

import tensorflow as tf
from tensorflow.keras.layers import Dense, LeakyReLU, Dropout, Input
from tensorflow.keras.models import Sequential, Model
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.initializers import RandomNormal
import numpy as np
import matplotlib.pyplot as plt

Hyperparameters

NOISE_DIM을 정의 합니다.

NOISE_DIM은 자유롭게 설정할 수 있으며, generator의 input으로 들어갑니다.

# gan에 입력되는 noise에 대한 dimension
NOISE_DIM = 10

# adam optimizer 정의, learning_rate = 0.0002, beta_1로 줍니다.
# Vanilla Gan과 DCGAN에서 이렇게 셋팅을 해주는데
# 이렇게 해줘야 훨씬 학습을 잘합니다.
adam = Adam(lr=0.0002, beta_1=0.5)

Generator

generator를 정의합니다.

generator = Sequential([
    Dense(256, input_dim=NOISE_DIM), 
    LeakyReLU(0.2), 
    Dense(512), 
    LeakyReLU(0.2), 
    Dense(1024), 
    LeakyReLU(0.2), 
    Dense(28*28, activation='tanh'),
])

generator.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                (None, 256)               2816      
_________________________________________________________________
leaky_re_lu (LeakyReLU)      (None, 256)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 512)               131584    
_________________________________________________________________
leaky_re_lu_1 (LeakyReLU)    (None, 512)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 1024)              525312    
_________________________________________________________________
leaky_re_lu_2 (LeakyReLU)    (None, 1024)              0         
_________________________________________________________________
dense_3 (Dense)              (None, 784)               803600    
=================================================================
Total params: 1,463,312
Trainable params: 1,463,312
Non-trainable params: 0
_________________________________________________________________

Discriminator

discriminator를 정의합니다.

discriminator = Sequential([
    Dense(1024, input_shape=(784,), kernel_initializer=RandomNormal(stddev=0.02)),
    LeakyReLU(0.2), 
    Dropout(0.3), 
    Dense(512),
    LeakyReLU(0.2), 
    Dropout(0.3), 
    Dense(256),
    LeakyReLU(0.2), 
    Dropout(0.3), 
    Dense(1, activation='sigmoid')
])

discriminator.summary()

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_4 (Dense)              (None, 1024)              803840    
_________________________________________________________________
leaky_re_lu_3 (LeakyReLU)    (None, 1024)              0         
_________________________________________________________________
dropout (Dropout)            (None, 1024)              0         
_________________________________________________________________
dense_5 (Dense)              (None, 512)               524800    
_________________________________________________________________
leaky_re_lu_4 (LeakyReLU)    (None, 512)               0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_6 (Dense)              (None, 256)               131328    
_________________________________________________________________
leaky_re_lu_5 (LeakyReLU)    (None, 256)               0         
_________________________________________________________________
dropout_2 (Dropout)          (None, 256)               0         
_________________________________________________________________
dense_7 (Dense)              (None, 1)                 257       
=================================================================
Total params: 1,460,225
Trainable params: 1,460,225
Non-trainable params: 0
_________________________________________________________________

반드시 dicriminator를 compile 해주어야 합니다.

discriminator.compile(loss='binary_crossentropy', optimizer=adam)

Gan

generator와 discriminator를 연결합니다.

# discriminator는 학습을 하지 않도록 하며, Gan 모델에서는 generator만 학습하도록 합니다.
discriminator.trainable = False
gan_input = Input(shape=(NOISE_DIM,))
x = generator(inputs=gan_input)
output = discriminator(x)

gan 모델을 정의합니다.

gan = Model(gan_input, output)

gan.summary()

Model: "model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         [(None, 10)]              0         
_________________________________________________________________
sequential (Sequential)      (None, 784)               1463312   
_________________________________________________________________
sequential_1 (Sequential)    (None, 1)                 1460225   
=================================================================
Total params: 2,923,537
Trainable params: 1,463,312
Non-trainable params: 1,460,225
_________________________________________________________________

Compile

gan.compile(loss='binary_crossentropy', optimizer=adam)

Batch

이미지 batch를 생성합니다. MNIST 이미지 batch가 차례대로 생성됩니다.

def get_batches(data, batch_size):
    batches = []
    for i in range(int(data.shape[0] // batch_size)):
        batch = data[i * batch_size: (i + 1) * batch_size]
        batches.append(batch)
    return np.asarray(batches)

시각화를 위한 유틸 함수 정의

def visualize_training(epoch, d_losses, g_losses):
    # 오차에 대한 시각화
    plt.figure(figsize=(8, 4))
    plt.plot(d_losses, label='Discriminator Loss')
    plt.plot(g_losses, label='Generatror Loss')
    plt.xlabel('Epoch')
    plt.ylabel('Loss')
    plt.legend()
    plt.show()
    
    print('epoch: {}, Discriminator Loss: {}, Generator Loss: {}'.format(epoch, np.asarray(d_losses).mean(), np.asarray(g_losses).mean()))
    
    #샘플 데이터 생성 후 시각화
    noise = np.random.normal(0, 1, size=(24, NOISE_DIM))
    generated_images = generator.predict(noise)
    generated_images = generated_images.reshape(-1, 28, 28)
    
    plt.figure(figsize=(8, 4))
    for i in range(generated_images.shape[0]):
        plt.subplot(4, 6, i+1)
        plt.imshow(generated_images[i], interpolation='nearest', cmap='gray')
        plt.axis('off')
    plt.tight_layout()
    plt.show()

학습

BATCH_SIZE = 128
EPOCHS= 50

loss_function = tf.keras.losses.BinaryCrossentropy()

optimizer = tf.keras.optimizers.Adam(lr=0.0002, beta_1=0.5)

train_loss = tf.keras.metrics.BinaryCrossentropy(name='train_loss')
train_accuracy = tf.keras.metrics.BinaryAccuracy(name='train_accuracy')

# discriminator와 gan 모델의 loss 측정을 위한 list 입니다.
d_losses = []
g_losses = []

for epoch in range(1, EPOCHS + 1):
    # 각 배치별 학습
    for real_images in get_batches(x_train, BATCH_SIZE):
        # 랜덤 노이즈 생성
        input_noise = np.random.uniform(-1, 1, size=[BATCH_SIZE, NOISE_DIM])
        
        # 가짜 이미지 데이터 생성
        generated_images = generator.predict(input_noise)
        
        # Gan에 학습할 X 데이터 정의
        x_dis = np.concatenate([real_images, generated_images])
        
        # Gan에 학습할 Y 데이터 정의
        y_dis = np.zeros(2 * BATCH_SIZE)
        y_dis[:BATCH_SIZE] = 0.9
        
        # Discriminator 훈련
        discriminator.trainable = True
        d_loss = discriminator.train_on_batch(x_dis, y_dis)
        
        # Gan 훈련
        noise = np.random.uniform(-1, 1, size=[BATCH_SIZE, NOISE_DIM])
        y_gan = np.ones(BATCH_SIZE)
        
        # Discriminator의 판별 학습을 방지합니다
        discriminator.trainable = False
        g_loss = gan.train_on_batch(noise, y_gan)
        
    d_losses.append(d_loss)
    g_losses.append(g_loss)

    if epoch == 1 or epoch % 5 == 0:
        visualize_training(epoch, d_losses, g_losses)

epoch: 1, Discriminator Loss: 0.5994292497634888, Generator Loss: 1.1322251558303833

epoch: 5, Discriminator Loss: 0.596705973148346, Generator Loss: 1.2031847953796386

epoch: 10, Discriminator Loss: 0.5923080146312714, Generator Loss: 1.231764554977417

epoch: 15, Discriminator Loss: 0.593469242254893, Generator Loss: 1.2403248151143391

epoch: 20, Discriminator Loss: 0.5942246615886688, Generator Loss: 1.2294869124889374

epoch: 25, Discriminator Loss: 0.5937843990325927, Generator Loss: 1.2424570941925048

epoch: 30, Discriminator Loss: 0.5932048420111339, Generator Loss: 1.2330509106318155

epoch: 35, Discriminator Loss: 0.5952658534049988, Generator Loss: 1.2308267593383788

epoch: 40, Discriminator Loss: 0.5956525504589081, Generator Loss: 1.230743545293808

epoch: 45, Discriminator Loss: 0.5950222147835625, Generator Loss: 1.22891116672092

epoch: 50, Discriminator Loss: 0.595591914653778, Generator Loss: 1.2263532972335816

epoch: 55, Discriminator Loss: 0.5951204711740667, Generator Loss: 1.2225543303923174

epoch: 60, Discriminator Loss: 0.5948738396167755, Generator Loss: 1.2203315754731496

epoch: 65, Discriminator Loss: 0.5955134492654067, Generator Loss: 1.2133508975689227

epoch: 70, Discriminator Loss: 0.5966339937278202, Generator Loss: 1.2099131516047887

epoch: 75, Discriminator Loss: 0.5964618261655171, Generator Loss: 1.2145421997706096

epoch: 80, Discriminator Loss: 0.5959978416562081, Generator Loss: 1.2144650369882584

epoch: 85, Discriminator Loss: 0.5979385621407453, Generator Loss: 1.210183084712309

epoch: 90, Discriminator Loss: 0.5982275896602207, Generator Loss: 1.2074756185213724

epoch: 95, Discriminator Loss: 0.5986613097943758, Generator Loss: 1.2066561623623497

epoch: 100, Discriminator Loss: 0.5988692629337311, Generator Loss: 1.2042953073978424

epoch: 105, Discriminator Loss: 0.5996920863787333, Generator Loss: 1.200414310182844

epoch: 110, Discriminator Loss: 0.6006603782827203, Generator Loss: 1.1974624883044849

epoch: 115, Discriminator Loss: 0.6003348262413688, Generator Loss: 1.1977796720421832

epoch: 120, Discriminator Loss: 0.6005971307555834, Generator Loss: 1.195457966128985

epoch: 125, Discriminator Loss: 0.6014634466171265, Generator Loss: 1.195823854446411

epoch: 130, Discriminator Loss: 0.6019030676438258, Generator Loss: 1.1958707176722012

epoch: 135, Discriminator Loss: 0.602101586483143, Generator Loss: 1.1991082527019359

epoch: 140, Discriminator Loss: 0.6023781512464795, Generator Loss: 1.1979700386524201

epoch: 145, Discriminator Loss: 0.6024142187217186, Generator Loss: 1.1975739289974343

epoch: 150, Discriminator Loss: 0.6024736766020456, Generator Loss: 1.1967066311836243

epoch: 155, Discriminator Loss: 0.6025569781180351, Generator Loss: 1.1936120625465148

epoch: 160, Discriminator Loss: 0.602316415309906, Generator Loss: 1.1947377875447274

epoch: 165, Discriminator Loss: 0.6024774937918692, Generator Loss: 1.1946386741869377

epoch: 170, Discriminator Loss: 0.6024079308790319, Generator Loss: 1.1935257161364836

epoch: 175, Discriminator Loss: 0.602534671170371, Generator Loss: 1.1943525968279158

epoch: 180, Discriminator Loss: 0.602770804034339, Generator Loss: 1.1952077547709148

epoch: 185, Discriminator Loss: 0.602991542945037, Generator Loss: 1.195974215945682

epoch: 190, Discriminator Loss: 0.6030043325926128, Generator Loss: 1.1957153370505884

epoch: 195, Discriminator Loss: 0.6031381720151656, Generator Loss: 1.197337226378612

epoch: 200, Discriminator Loss: 0.6032074254751205, Generator Loss: 1.1968027561903

epoch: 205, Discriminator Loss: 0.6029357628124516, Generator Loss: 1.1968888352556926

epoch: 210, Discriminator Loss: 0.6027095303648994, Generator Loss: 1.1974678641273862

epoch: 215, Discriminator Loss: 0.6023930959923323, Generator Loss: 1.1977701874666435

epoch: 220, Discriminator Loss: 0.6027434256943789, Generator Loss: 1.1969362589445982

epoch: 225, Discriminator Loss: 0.6026199438836839, Generator Loss: 1.1962328555848862

epoch: 230, Discriminator Loss: 0.6024436030698859, Generator Loss: 1.1960520277852598

epoch: 235, Discriminator Loss: 0.6027739088586036, Generator Loss: 1.1950260811663689

epoch: 240, Discriminator Loss: 0.602807242423296, Generator Loss: 1.1951396644115448

epoch: 245, Discriminator Loss: 0.6032830462163808, Generator Loss: 1.1955129749920903

epoch: 250, Discriminator Loss: 0.6033235461711883, Generator Loss: 1.1946135258674622

epoch: 255, Discriminator Loss: 0.6035783564343172, Generator Loss: 1.194843908852222

epoch: 260, Discriminator Loss: 0.6035869275148098, Generator Loss: 1.1946847975254058

epoch: 265, Discriminator Loss: 0.60363453154294, Generator Loss: 1.1940310325262682

epoch: 270, Discriminator Loss: 0.6038364982163464, Generator Loss: 1.1934597421575475

epoch: 275, Discriminator Loss: 0.6039837902242488, Generator Loss: 1.193264699415727

epoch: 280, Discriminator Loss: 0.6038310159529958, Generator Loss: 1.1932028361729212

epoch: 285, Discriminator Loss: 0.6036519828595613, Generator Loss: 1.1925087108946684

epoch: 290, Discriminator Loss: 0.603931030733832, Generator Loss: 1.1912308532616187

epoch: 295, Discriminator Loss: 0.6038557482978045, Generator Loss: 1.1915957644834356

epoch: 300, Discriminator Loss: 0.6036268909772237, Generator Loss: 1.1933002404371897