10/30/24, 8:33 PM 4PA21_Aliyah Muthi Lathifah_10521097 - Colab
keyboard_arrow_down 1) Importing Python Packages for GAN
from keras.datasets import cifar10, mnist
from keras.models import Sequential
from keras.layers import Reshape
from keras.layers import Dense
from keras.layers import Flatten
from keras.layers import Conv2D
from keras.layers import Conv2DTranspose
from keras.layers import Dropout
from keras.layers import LeakyReLU
from tensorflow.keras.optimizers import Adam
import numpy as np
!mkdir generated_images
keyboard_arrow_down 2) Parameters for Neural Networks & Data
img_width = 32
img_height = 32
channels = 3
img_shape = (img_width, img_height, channels)
latent_dim = 100
adam = Adam(learning_rate=0.0002)
keyboard_arrow_down 3) Building Generator
def build_generator():
model = Sequential()
# Create first layer, to receive the input
model.add(Dense(256 * 4 * 4, input_dim = latent_dim))
# 256 * 8 * 8; for upscaling the layers,
# initial shape to construct into final shape
# Create default activation function
model.add(LeakyReLU(alpha = 0.2))
# Create reshape layer
model.add(Reshape((4, 4,256)))
# 8,8,256 ; reffers to first layer
# Adding more layers for neurons and better result
model.add(Conv2DTranspose(128, (4,4), strides = (2,2), padding = 'same'))
model.add(LeakyReLU(alpha= 0.2))
model.add(Conv2DTranspose(128, (4,4), strides = (2,2), padding = 'same'))
model.add(LeakyReLU(alpha= 0.2))
model.add(Conv2DTranspose(128, (4,4), strides = (2,2), padding = 'same'))
model.add(LeakyReLU(alpha= 0.2))
# (4,4) >> filter size
# strides = (2,2) >> Convolutional layers, that how NN understand images
# Create Final output layer and forming image shape
# the shape (3, (3,3)) reffers to image shape :
# >>> img_shape = (img_width, img_height, channels)
model.add(Conv2D(3, (3,3), activation= 'tanh', padding = 'same'))
#
model.summary()
return model
generator = build_generator()
keyboard_arrow_down 4) Building Discriminator
def build_discriminator():
model = Sequential()
# Create input layer and filter and stride layer. That makes NN understand image
model.add(Conv2D(64, (3,3), padding = 'same', input_shape = img_shape))
# Adding activation function
https://colab.research.google.com/drive/1vnJnWO16rcVPpV5yEacL0XFQ6styQ-_A#printMode=true 1/4
10/30/24, 8:33 PM 4PA21_Aliyah Muthi Lathifah_10521097 - Colab
model.add(LeakyReLU(alpha = 0.2))
model.add(Conv2D(128, (3,3), padding = 'same'))
model.add(LeakyReLU(alpha = 0.2))
model.add(Conv2D(128, (3,3), padding = 'same'))
model.add(LeakyReLU(alpha = 0.2))
model.add(Conv2D(256, (3,3), padding = 'same'))
model.add(LeakyReLU(alpha = 0.2))
model.add(Flatten())
model.add(Dropout(0.4))
# Create output layer
model.add(Dense(1, activation = 'sigmoid'))
model.summary()
return model
discriminator = build_discriminator()
discriminator.compile(loss='binary_crossentropy', optimizer=adam, metrics=['accuracy'])
keyboard_arrow_down 5) Connecting Neural Networks to build GAN
GAN = Sequential()
discriminator.trainable = False
GAN.add(generator)
GAN.add(discriminator)
GAN.compile(loss='binary_crossentropy', optimizer=adam)
GAN.summary()
Model: "sequential_2"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ Layer (type) ┃ Output Shape ┃ Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ sequential (Sequential) │ (None, 32, 32, 3) │ 1,466,115 │
├──────────────────────────────────────┼─────────────────────────────┼─────────────────┤
│ sequential_1 (Sequential) │ (None, 1) │ 780,545 │
└──────────────────────────────────────┴─────────────────────────────┴─────────────────┘
Total params: 2,246,660 (8.57 MB)
Trainable params: 1,466,115 (5.59 MB)
Non-trainable params: 780,545 (2.98 MB)
keyboard_arrow_down 6) Outputting Images
import matplotlib.pyplot as plt
import glob
import imageio
import PIL
save_name = 0.00000000
def save_imgs(epoch):
r, c = 5, 5
noise = np.random.normal(0, 1, (r * c, latent_dim))
gen_imgs = generator.predict(noise)
global save_name
save_name += 0.00000001
# print("%.8f" % save_name)
# Rescale images 0 - 1
# gen_imgs = 0.5 * gen_imgs + 0.5
gen_imgs = (gen_imgs + 1) / 2.0
# gen_imgs = gen_imgs * 255
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])
axs[i,j].axis('off')
cnt += 1
fig.savefig("generated_images/%.8f.png" % save_name)
plt.close()
https://colab.research.google.com/drive/1vnJnWO16rcVPpV5yEacL0XFQ6styQ-_A#printMode=true 2/4
10/30/24, 8:33 PM 4PA21_Aliyah Muthi Lathifah_10521097 - Colab
keyboard_arrow_down 7) Training GAN
def train(epochs, batch_size = 64, save_interval = 200):
(X_train, _), (_, _) = cifar10.load_data()
# Rescaling the data
X_train = X_train / 127.5 -1.
bat_per_epo = int(X_train.shape[0] / batch_size)
# Create Y label for NN
valid = np.ones((batch_size,1))
fakes = np.zeros((batch_size, 1))
for epoch in range (epochs) :
for j in range(bat_per_epo) :
#Get Random Batch
idx = np.random.randint(0, X_train.shape[0], batch_size)
imgs = X_train[idx]
# Generate Fakes Images
noise = np.random.normal(0, 1, (batch_size, latent_dim))
gen_imgs = generator.predict(noise)
# Train Discriminator
d_loss_real = discriminator.train_on_batch(imgs, valid)
d_loss_fake = discriminator.train_on_batch(gen_imgs, fakes)
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
noise = np.random.normal(0, 1, (batch_size, latent_dim))
# Inverse Y label
g_loss = GAN.train_on_batch(noise, valid)
# Assuming d_loss and g_loss are lists containing loss and accuracy:
# print('******* %d %d [D loss: %f, acc: %.2f%%] [G loss: %f]' % (epoch, d_loss[0],100*d_loss[1], g_loss))
print("******* %d [D loss: %f, acc: %.2f%%] [G loss: %f]" % (epoch, d_loss[0], 100* d_loss[1], g_loss[0])) # Access the first elem
save_imgs(epoch)
train(1000, batch_size = 64, save_interval = 200)
noise = np.random.normal(0, 1, (1,latent_dim))
gen_imgs = generator.predict(noise)
gen_imgs = (gen_imgs + 1) / 2.0
gen_imgs = gen_imgs.squeeze(0)
plt.imshow(gen_imgs)
keyboard_arrow_down 8) Making GIF
# Display a single image using the epoch number
# def display_image(epoch_no):
# return PIL.Image.open('generated_images/%.8f.png'.format(epoch_no))
anim_file = 'dcgan.gif'
with imageio.get_writer(anim_file, mode='I') as writer:
filenames = glob.glob('generated_images/*.png')
filenames = sorted(filenames)
for filename in filenames:
image = imageio.imread(filename)
writer.append_data(image)
#The if statement was not indented at the correct level
if filenames: # Check if filenames list is not empty
last_filename = filenames[-1]
image = imageio.imread(last_filename)
writer.append_data(image)
https://colab.research.google.com/drive/1vnJnWO16rcVPpV5yEacL0XFQ6styQ-_A#printMode=true 3/4
10/30/24, 8:33 PM 4PA21_Aliyah Muthi Lathifah_10521097 - Colab
https://colab.research.google.com/drive/1vnJnWO16rcVPpV5yEacL0XFQ6styQ-_A#printMode=true 4/4
