{"id":67054,"date":"2024-06-17T17:04:43","date_gmt":"2024-06-17T13:34:43","guid":{"rendered":"https:\/\/nabfollower.com\/blog\/from-noise-to-art-building-your-first-generative-adversarial-network-472o\/"},"modified":"2024-06-17T17:04:43","modified_gmt":"2024-06-17T13:34:43","slug":"from-noise-to-art-building-your-first-generative-adversarial-network-472o","status":"publish","type":"post","link":"https:\/\/nabfollower.com\/blog\/from-noise-to-art-building-your-first-generative-adversarial-network-472o\/","title":{"rendered":"\u0627\u0632 \u0646\u0648\u06cc\u0632 \u062a\u0627 \u0647\u0646\u0631: \u0633\u0627\u062e\u062a\u0646 \u0627\u0648\u0644\u06cc\u0646 \u0634\u0628\u06a9\u0647 \u0645\u062a\u062e\u0627\u0635\u0645 \u0645\u0648\u0644\u062f \u0634\u0645\u0627"},"content":{"rendered":"

Summarize this content to 400 words in Persian Lang
\n \u0645\u0646 \u0628\u0627 \u0627\u06cc\u0646 \u0627\u06cc\u062f\u0647 \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0645\u0627\u0634\u06cc\u0646\u06cc \u0628\u0627\u0634\u06a9\u0648\u0647 \u06a9\u0647 \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0634\u0628\u06a9\u0647\u200c\u0647\u0627\u06cc \u0645\u062a\u062e\u0627\u0635\u0645 \u0645\u0648\u0644\u062f (GAN) \u0634\u0646\u0627\u062e\u062a\u0647 \u0645\u06cc\u200c\u0634\u0648\u062f\u060c \u0628\u0647 \u062e\u0635\u0648\u0635 \u062f\u0631 \u062d\u0648\u0632\u0647 \u062a\u0648\u0644\u06cc\u062f \u062a\u0635\u0648\u06cc\u0631 \u0622\u0634\u0646\u0627 \u0634\u062f\u0645. \u0686\u0627\u0631\u0686\u0648\u0628 \u062f\u06cc\u06af\u0631\u06cc \u0628\u0647 \u0646\u0627\u0645 GANs \u062a\u0648\u0633\u0637 \u0627\u06cc\u0627\u0646 \u06af\u0648\u062f\u0641\u0644\u0648 \u062f\u0631 \u0633\u0627\u0644 2014 \u062a\u0648\u0633\u0639\u0647 \u06cc\u0627\u0641\u062a. \u0645\u0639\u0645\u0627\u0631\u06cc \u0632\u06cc\u0631\u0628\u0646\u0627\u06cc\u06cc \u0622\u0646 \u0628\u0627 \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0627\u0632 \u0631\u0642\u0627\u0628\u062a \u062f\u0648 \u0634\u0628\u06a9\u0647 \u0639\u0635\u0628\u06cc \u0633\u0627\u062e\u062a\u0647 \u0634\u062f\u0647 \u0627\u0633\u062a. \u0628\u0627 \u062a\u0648\u062c\u0647 \u0628\u0647 \u0648\u0633\u0639\u062a \u0627\u06cc\u0646 \u0648\u0628\u0644\u0627\u06af\u060c \u0627\u062c\u0627\u0632\u0647 \u062f\u0647\u06cc\u062f \u0627\u0628\u062a\u062f\u0627 \u0645\u0639\u0631\u0641\u06cc \u06a9\u0646\u0645 GAN \u0686\u06cc\u0633\u062a\u060c \u0648 \u0633\u067e\u0633 \u0628\u0647 \u0634\u0645\u0627 \u0628\u06af\u0648\u06cc\u0645 \u06a9\u0647 \u062f\u0631 \u0627\u06cc\u0646 \u0648\u0628\u0644\u0627\u06af \u0627\u0632 \u062c\u0645\u0644\u0647 \u06a9\u062f \u0645\u0648\u062c\u0648\u062f \u062f\u0631 TensorFlow \u062f\u0631 \u0645\u0648\u0631\u062f \u0646\u062d\u0648\u0647 \u0622\u0645\u0648\u0632\u0634 \u06cc\u06a9 GAN \u0633\u0627\u062f\u0647\u060c \u0686\u0647 \u06a9\u0627\u0631 \u062e\u0648\u0627\u0647\u0645 \u06a9\u0631\u062f.<\/p>\n

GAN \u0686\u06cc\u0633\u062a\u061f\u062f\u0631 \u0647\u0633\u062a\u0647 \u062e\u0648\u062f\u060c \u06cc\u06a9 GAN \u0627\u0632 \u062f\u0648 \u0634\u0628\u06a9\u0647 \u0639\u0635\u0628\u06cc \u062a\u0634\u06a9\u06cc\u0644 \u0634\u062f\u0647 \u0627\u0633\u062a: \u0627\u0644\u0628\u062a\u0647\u060c \u0645\u0648\u0644\u062f \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u062c\u0639\u0644\u06cc \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f\u060c \u0648 \u062a\u0645\u0627\u06cc\u0632 \u062f\u0647\u0646\u062f\u0647 \u0627\u06cc \u06a9\u0647 \u06cc\u0627\u062f \u0645\u06cc \u06af\u06cc\u0631\u062f \u0686\u06af\u0648\u0646\u0647 \u0628\u06cc\u0646 \u0686\u06cc\u0632\u0647\u0627\u06cc \u062c\u0639\u0644\u06cc \u0648 \u0648\u0627\u0642\u0639\u06cc \u062a\u0645\u0627\u06cc\u0632 \u0642\u0627\u0626\u0644 \u0634\u0648\u062f. <\/p>\n

Generator: \u067e\u0633 \u0627\u0632 \u0648\u0627\u0631\u062f \u06a9\u0631\u062f\u0646 \u0646\u0648\u06cc\u0632 \u0648 \u0633\u067e\u0633 \u0639\u0628\u0648\u0631 \u062f\u0627\u062f\u0646 \u0622\u0646\u0647\u0627 \u0628\u0631\u0627\u06cc \u062a\u0648\u0644\u06cc\u062f \u06cc\u06a9 \u062f\u0627\u062f\u0647 \u062e\u0631\u0648\u062c\u06cc \u06a9\u0647 \u0634\u0628\u06cc\u0647 \u0627\u0644\u06af\u0648\u06cc \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc \u0627\u0633\u062a.
\n\u062a\u0641\u06a9\u06cc\u06a9 \u06a9\u0646\u0646\u062f\u0647: \u062a\u0645\u0627\u06cc\u0632 \u06a9\u0646\u0646\u062f\u0647 \u0628\u0647 \u06a9\u0627\u0631 \u06af\u0631\u0641\u062a\u0647 \u0634\u062f\u0647 \u062f\u0631 \u062a\u0648\u0635\u06cc\u0641 \u0627\u06cc\u0646 \u0645\u062f\u0644 \u06cc\u06a9 \u0646\u0645\u0648\u0646\u0647 \u0648\u0631\u0648\u062f\u06cc \u0645\u06cc \u06af\u06cc\u0631\u062f \u0648 \u0633\u0639\u06cc \u0645\u06cc \u06a9\u0646\u062f \u062d\u062f\u0633 \u0628\u0632\u0646\u062f \u06a9\u0647 \u0622\u06cc\u0627 \u0646\u0645\u0648\u0646\u0647 \u0627\u0632 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc \u06af\u0631\u0641\u062a\u0647 \u0634\u062f\u0647 \u0627\u0633\u062a \u06cc\u0627 \u0641\u0642\u0637 \u0628\u0627 \u06a9\u0645\u06a9 \u0698\u0646\u0631\u0627\u062a\u0648\u0631 \u0633\u0646\u062a\u0632 \u0634\u062f\u0647 \u0627\u0633\u062a.<\/p>\n

\u0627\u06cc\u0646 \u062f\u0648 \u0634\u0628\u06a9\u0647 \u0628\u0647 \u0637\u0648\u0631 \u0647\u0645\u0632\u0645\u0627\u0646 \u062f\u0631 \u06cc\u06a9 \u0686\u0627\u0631\u0686\u0648\u0628 \u0628\u0627\u0632\u06cc \u062d\u0627\u0635\u0644 \u062c\u0645\u0639 \u0635\u0641\u0631 \u0622\u0645\u0648\u0632\u0634 \u062f\u0627\u062f\u0647 \u0645\u06cc \u0634\u0648\u0646\u062f: \u062f\u0631 \u062d\u0627\u0644\u06cc \u06a9\u0647 \u062f\u0631 GAN \u0647\u0627\u060c \u0634\u0628\u06a9\u0647 \u0645\u0648\u0644\u062f \u0627\u0637\u0644\u0627\u0639\u0627\u062a \u0631\u0627 \u0628\u0647 \u0634\u0628\u06a9\u0647 \u0645\u062a\u0645\u0627\u06cc\u0632 \u0645\u06cc \u062f\u0647\u062f \u062a\u0627 \u0622\u0646 \u0631\u0627 \u0641\u0631\u06cc\u0628 \u062f\u0647\u062f \u062a\u0627 \u0628\u0627\u0648\u0631 \u06a9\u0646\u062f \u06a9\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u062f\u0627\u062f\u0647 \u0634\u062f\u0647 \u0628\u0647 \u0622\u0646 \u0648\u0627\u0642\u0639\u06cc \u0627\u0633\u062a\u060c \u0627\u0645\u0627 \u0627\u0632 \u0637\u0631\u0641 \u062f\u06cc\u06af\u0631\u060c \u0627\u0637\u0644\u0627\u0639\u0627\u062a \u0645\u062a\u0645\u0627\u06cc\u0632 \u06a9\u0646\u0646\u062f\u0647 \u0627\u0633\u062a. \u0634\u0628\u06a9\u0647 \u0646\u0642\u0634 \u062a\u0634\u062e\u06cc\u0635 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0648\u0627\u0642\u0639\u06cc \u0627\u0632 \u062c\u0639\u0644\u06cc \u0631\u0627 \u062f\u0627\u0631\u062f. <\/p>\n

\u0631\u0627\u0647\u0646\u0645\u0627\u06cc \u06af\u0627\u0645 \u0628\u0647 \u06af\u0627\u0645 \u0633\u0627\u062e\u062a \u06cc\u06a9 GAN \u0633\u0627\u062f\u0647<\/p>\n

\u0645\u0631\u062d\u0644\u0647 1: \u062a\u0646\u0638\u06cc\u0645 \u0645\u062d\u06cc\u0637<\/p>\n

pip install tensorflow<\/p>\n

\u0645\u0631\u062d\u0644\u0647 2: \u0648\u0627\u0631\u062f \u06a9\u0631\u062f\u0646 \u06a9\u062a\u0627\u0628\u062e\u0627\u0646\u0647 \u0647\u0627\u06cc \u0636\u0631\u0648\u0631\u06cc<\/p>\n

import tensorflow as tf
\nfrom tensorflow.keras import layers
\nimport numpy as np
\nimport matplotlib.pyplot as plt<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

\u0645\u0631\u062d\u0644\u0647 3: \u0698\u0646\u0631\u0627\u062a\u0648\u0631 \u0631\u0627 \u062a\u0639\u0631\u06cc\u0641 \u06a9\u0646\u06cc\u062f<\/p>\n

\u0634\u0628\u06a9\u0647 \u0645\u0648\u0644\u062f \u0633\u067e\u0633 \u06cc\u06a9 \u0628\u0631\u062f\u0627\u0631 \u0646\u0648\u06cc\u0632 \u0628\u0647 \u0637\u0648\u0631 \u062a\u0635\u0627\u062f\u0641\u06cc \u0627\u0646\u062a\u062e\u0627\u0628 \u0645\u06cc \u06a9\u0646\u062f \u0648 \u0622\u0646 \u0631\u0627 \u062f\u0631 \u06cc\u06a9 \u0646\u0642\u0637\u0647 \u062f\u0627\u062f\u0647 \u06a9\u0647 \u0634\u0628\u06cc\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc \u0648\u0627\u0642\u0639\u06cc \u0627\u0633\u062a\u060c \u062a\u0631\u0633\u06cc\u0645 \u0645\u06cc \u06a9\u0646\u062f.<\/p>\n

def build_generator():
\n model = tf.keras.Sequential()
\n model.add(layers.Dense(8*8*128, use_bias=False, input_shape=(100,)))
\n model.add(layers.BatchNormalization())
\n model.add(layers.LeakyReLU())
\n model.add(layers.Reshape((8, 8, 128)))
\n assert model.output_shape == (None, 8, 8, 128) # Note: None is the batch size<\/p>\n

model.add(layers.Conv2DTranspose(128, (5, 5), strides=(1, 1), padding=’same’, use_bias=False))
\n assert model.output_shape == (None, 8, 8, 128)
\n model.add(layers.BatchNormalization())
\n model.add(layers.LeakyReLU())<\/p>\n

model.add(layers.Conv2DTranspose(128, (5, 5), strides=(2, 2), padding=’same’, use_bias=False))
\n model.add(layers.BatchNormalization())
\n model.add(layers.LeakyReLU())
\n assert model.output_shape == (None, 16, 16, 128)<\/p>\n

model.add(layers.Conv2DTranspose(128, (5, 5), strides=(2, 2), padding=’same’, use_bias=False))
\n model.add(layers.BatchNormalization())
\n model.add(layers.LeakyReLU())
\n assert model.output_shape == (None, 32, 32, 128)<\/p>\n

model.add(layers.Conv2DTranspose(128, (5, 5), strides=(2, 2), padding=’same’, use_bias=False))
\n model.add(layers.BatchNormalization())
\n model.add(layers.LeakyReLU())
\n assert model.output_shape == (None, 64, 64, 128)<\/p>\n

model.add(layers.Conv2DTranspose(3, (5, 5), strides=(2, 2), padding=’same’, use_bias=False, activation=’tanh’))
\n print(model.output_shape)<\/p>\n

return model<\/p>\n

generator = build_generator()
\ngenerator.summary()<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

\u0645\u0631\u062d\u0644\u0647 4: \u062a\u0634\u062e\u06cc\u0635 \u062f\u0647\u0646\u062f\u0647 \u0631\u0627 \u062a\u0639\u0631\u06cc\u0641 \u06a9\u0646\u06cc\u062f<\/p>\n

\u0634\u0628\u06a9\u0647 \u062a\u0634\u062e\u06cc\u0635 \u062f\u0647\u0646\u062f\u0647 \u06cc\u06a9 \u0646\u0645\u0648\u0646\u0647 \u0648\u0631\u0648\u062f\u06cc \u0631\u0627 \u0645\u06cc \u06af\u06cc\u0631\u062f \u0648 \u0622\u0646 \u0631\u0627 \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0648\u0627\u0642\u0639\u06cc \u0637\u0628\u0642\u0647 \u0628\u0646\u062f\u06cc \u0645\u06cc \u06a9\u0646\u062f<\/p>\n

def build_discriminator():
\n model = tf.keras.Sequential()
\n model.add(layers.Conv2D(64, (5, 5), strides=(2, 2), padding=’same’,
\n input_shape=[128, 128, 3]))
\n model.add(layers.LeakyReLU())
\n model.add(layers.Dropout(0.3))<\/p>\n

model.add(layers.Conv2D(128, (5, 5), strides=(2, 2), padding=’same’))
\n model.add(layers.LeakyReLU())
\n model.add(layers.Dropout(0.3))<\/p>\n

model.add(layers.Conv2D(256, (5, 5), strides=(2, 2), padding=’same’))
\n model.add(layers.LeakyReLU())
\n model.add(layers.Dropout(0.3))<\/p>\n

model.add(layers.Flatten())
\n model.add(layers.Dense(1))
\n return model<\/p>\n

discriminator = build_discriminator()
\ndiscriminator.summary()<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

\u0645\u0631\u062d\u0644\u0647 5: \u0645\u062f\u0644 \u0647\u0627 \u0631\u0627 \u062a\u0633\u062a \u06a9\u0646\u06cc\u062f<\/p>\n

noise = tf.random.normal([1,100])
\ngenerated_image = generator(noise,training=False)
\nprint(discriminator(generated_image))
\nplt.imshow(generated_image[0]*127.5+127.5)<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

\u0645\u0631\u062d\u0644\u0647 6: \u0631\u0627\u0647 \u0627\u0646\u062f\u0627\u0632\u06cc \u0639\u0645\u0644\u06a9\u0631\u062f \u0627\u0632 \u062f\u0633\u062a \u062f\u0627\u062f\u0646 \u0648 \u0628\u0647\u06cc\u0646\u0647 \u0633\u0627\u0632<\/p>\n

cross_entropy=BinaryCrossentropy(from_logits=True)<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

def discriminator_loss(real_output,fake_output):
\n real_loss = cross_entropy(tf.ones_like(real_output),real_output)
\n fake_loss = cross_entropy(tf.zeros_like(fake_output),fake_output)
\n total_loss = real_loss + fake_loss
\n return total_loss<\/p>\n

def generator_loss(fake_output):
\n return cross_entropy(tf.ones_like(fake_output),fake_output)<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

generator_optimizer = tf.keras.optimizers.Adam(1e-4)
\ndiscriminator_optimizer = tf.keras.optimizers.Adam(1e-4)<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

\u0645\u0631\u062d\u0644\u0647 7: \u0631\u0627\u0647 \u0627\u0646\u062f\u0627\u0632\u06cc \u0627\u06cc\u0633\u062a \u0628\u0627\u0632\u0631\u0633\u06cc<\/p>\n

checkpoint_dir=”training_checkpoints”
\ncheckpoint_prefix = os.path.join(checkpoint_dir,’ckpt’)
\ncheckpoint = tf.train.Checkpoint(generator_optimizer=generator_optimizer,
\n discriminator_optimizer=discriminator_optimizer,
\n generator=generator,
\n discriminator=discriminator)<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

\u0645\u0631\u062d\u0644\u0647 8: \u062a\u0639\u0631\u06cc\u0641 \u0645\u0631\u062d\u0644\u0647 \u0642\u0637\u0627\u0631<\/p>\n

@tf.function
\ndef train_step(images):<\/p>\n

noise=tf.random.normal([batch_size,noise_dims])<\/p>\n

with tf.GradientTape() as gen_tape, tf.GradientTape() as dis_tape:
\n generated_images=generator(noise,training=True)<\/p>\n

real_output=discriminator(images,training=True)
\n fake_output=discriminator(generated_images,training=True)<\/p>\n

gen_loss=generator_loss(fake_output)
\n disc_loss=discriminator_loss(real_output,fake_output)<\/p>\n

gen_gradients=gen_tape.gradient(gen_loss,generator.trainable_variables)
\n dis_gradients=dis_tape.gradient(disc_loss,discriminator.trainable_variables)<\/p>\n

generator_optimizer.apply_gradients(zip(gen_gradients,generator.trainable_variables))
\n discriminator_optimizer.apply_gradients(zip(dis_gradients,discriminator.trainable_variables))<\/p>\n

return gen_loss,disc_loss<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

\u0645\u0631\u062d\u0644\u0647 9: \u0631\u0627\u0647 \u0627\u0646\u062f\u0627\u0632\u06cc \u062d\u0644\u0642\u0647 \u0622\u0645\u0648\u0632\u0634\u06cc \u0648 \u0630\u062e\u06cc\u0631\u0647 \u062a\u0635\u0627\u0648\u06cc\u0631 \u062a\u0648\u0644\u06cc\u062f \u0634\u062f\u0647<\/p>\n

from IPython import display
\nimport time<\/p>\n

total_gloss=[]\ntotal_dloss=[]\ndef train(dataset,epochs):
\n for epoch in range(epochs):
\n disc_loss=gen_loss=0
\n start=time.time()
\n count=0
\n for batch in dataset:
\n losses=train_step(batch)
\n count+=1
\n disc_loss+=losses[1]\n gen_loss+=losses[0]\n total_gloss.append(gen_loss.numpy())
\n total_dloss.append(disc_loss.numpy())<\/p>\n

if (epoch+1)%50==0:
\n checkpoint.save(file_prefix=checkpoint_prefix)
\n display.clear_output(wait=True)
\n generate_and_save_output(generator,epoch+1,seed)<\/p>\n

print(f’Time for epoch {epoch + 1} is {time.time()-start}’)
\n print(f’Gloss: {gen_loss.numpy()\/count} , Dloss: {disc_loss.numpy()\/count}’,end=’\\n\\n’)
\n display.clear_output(wait=True)
\n generate_and_save_output(generator,epochs,seed)<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

def generate_and_save_output(model,epoch,test_input):<\/p>\n

predictions = model(test_input,training=False)
\n fig = plt.figure(figsize=(4,4))
\n for i in range(predictions.shape[0]):
\n plt.subplot(4,4,i+1)
\n plt.imshow((predictions[i]*127.5+127.5).numpy().astype(np.uint8),cmap=’gray’)
\n plt.axis(‘off’)
\n plt.savefig(f’image_at_epoch_{epoch}.png’)
\n plt.show()<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

\u0645\u0631\u062d\u0644\u0647 10: GAN \u0631\u0627 \u0622\u0645\u0648\u0632\u0634 \u062f\u0647\u06cc\u062f<\/p>\n

\u0628\u06cc\u0627\u06cc\u06cc\u062f GAN \u062e\u0648\u062f \u0631\u0627 \u0622\u0645\u0648\u0632\u0634 \u062f\u0647\u06cc\u0645\u060c \u0645\u0646 \u0627\u0632 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u062a\u0635\u0648\u06cc\u0631 \u0633\u06af \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u06a9\u0631\u062f\u0647 \u0627\u0645\u060c \u06a9\u0647 \u062f\u0631 \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0633\u06af \u0627\u0633\u062a\u0627\u0646\u0641\u0648\u0631\u062f Kaggle \u0645\u0648\u062c\u0648\u062f \u0627\u0633\u062a<\/p>\n

EPOCHS = 500
\nnoise_dims = 100
\nnum_egs_to_generate = 16
\nseed = tf.random.normal([num_egs_to_generate,noise_dims])<\/p>\n

train(train_images,EPOCHS)<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

\u062a\u0648\u062c\u0647: \u0628\u0631\u0627\u06cc \u062a\u0648\u0644\u06cc\u062f \u062a\u0635\u0627\u0648\u06cc\u0631 \u0628\u0627 \u06a9\u06cc\u0641\u06cc\u062a \u062e\u0648\u0628\u060c \u0645\u062f\u0644 \u0628\u0647 \u062a\u0639\u062f\u0627\u062f \u0632\u06cc\u0627\u062f\u06cc \u062f\u0648\u0631\u0647 \u0646\u06cc\u0627\u0632 \u062f\u0627\u0631\u062f.<\/p>\n

\u062f\u0631 \u062d\u0627\u0644 \u0627\u0645\u062a\u062d\u0627\u0646 \u06a9\u0631\u062f\u0646 \u0645\u062f\u0644 \u0645\u0627:<\/p>\n

new_image = generator(tf.random.normal([1,100]),training=False)
\nplt.imshow((new_image[0]*127.5+127.5).numpy().astype(np.uint8))<\/p>\n

\u0648\u0627\u0631\u062f \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u0634\u0648\u06cc\u062f<\/p>\n

\u0627\u0632 \u062d\u0627\u0644\u062a \u062a\u0645\u0627\u0645 \u0635\u0641\u062d\u0647 \u062e\u0627\u0631\u062c \u0634\u0648\u06cc\u062f<\/p>\n

\u0646\u062a\u06cc\u062c\u0647GAN \u0647\u0627 \u062f\u0631 \u062a\u0648\u0644\u06cc\u062f \u0645\u062c\u0645\u0648\u0639\u0647 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0648\u0627\u0642\u0639\u06cc \u0645\u0641\u06cc\u062f \u0647\u0633\u062a\u0646\u062f \u0632\u06cc\u0631\u0627 \u0622\u0646\u0647\u0627 \u0646\u0648\u0639\u06cc \u0634\u0628\u06a9\u0647 \u0639\u0635\u0628\u06cc \u0647\u0633\u062a\u0646\u062f \u06a9\u0647 \u0627\u0632 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u0622\u0645\u0648\u0632\u0634\u06cc \u0628\u0631\u0686\u0633\u0628 \u06af\u0630\u0627\u0631\u06cc \u0634\u062f\u0647 \u06cc\u0627\u062f \u0645\u06cc \u06af\u06cc\u0631\u0646\u062f \u0648 \u0633\u067e\u0633 \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u062c\u062f\u06cc\u062f\u06cc \u0627\u06cc\u062c\u0627\u062f \u0645\u06cc \u06a9\u0646\u0646\u062f. \u0627\u0632 \u0627\u06cc\u0646\u062c\u0627\u060c \u0641\u0631\u0645\u0648\u0644 \u0628\u0646\u062f\u06cc \u06cc\u06a9 GAN \u0645\u0639\u0642\u0648\u0644 \u0631\u0648\u0634\u0646 \u0648 \u0627\u0645\u06a9\u0627\u0646 \u067e\u0630\u06cc\u0631 \u062e\u0648\u0627\u0647\u062f \u0628\u0648\u062f \u0648 \u0627\u0632 \u0627\u06cc\u0646\u062c\u0627 \u0645\u0634\u0647\u0648\u062f \u0627\u0633\u062a \u06a9\u0647 \u06cc\u06a9 \u0631\u06cc\u062a\u0645 \u0646\u0633\u0628\u06cc \u0628\u06cc\u0646 \u062d\u0631\u06a9\u0627\u062a \u0645\u0648\u0644\u062f \u0648 \u0647\u0645\u0686\u0646\u06cc\u0646 \u062a\u0634\u062e\u06cc\u0635 \u062f\u0647\u0646\u062f\u0647 \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f. \u0627\u06cc\u0646 \u0627\u0645\u0631 \u0647\u062f\u0641 \u0631\u0627\u0647\u0646\u0645\u0627\u06cc \u06a9\u0646\u0648\u0646\u06cc \u0631\u0627 \u06a9\u0647 \u0635\u0631\u0641\u0627\u064b \u062e\u0648\u0627\u0646\u0646\u062f\u0647 \u0631\u0627 \u0628\u0627 \u0645\u0648\u0636\u0648\u0639 GAN \u0622\u0634\u0646\u0627 \u0645\u06cc\u200c\u06a9\u0646\u062f \u0648 \u0628\u0631\u0627\u06cc \u0627\u0648\u0644\u06cc\u0646 \u0628\u0627\u0631 \u0627\u0632 \u0622\u0646\u0686\u0647 \u062f\u0631 \u0627\u06cc\u0646 \u062d\u0648\u0632\u0647 \u062a\u062d\u0642\u06cc\u0642\u0627\u062a\u06cc \u0631\u0648 \u0628\u0647 \u0631\u0634\u062f \u0627\u0645\u06a9\u0627\u0646\u200c\u067e\u0630\u06cc\u0631 \u0627\u0633\u062a\u060c \u0628\u0647 \u0622\u0646\u0647\u0627 \u0627\u0631\u0627\u0626\u0647 \u0645\u06cc\u200c\u06a9\u0646\u062f. <\/p>\n

\u0645\u0646\u0627\u0628\u0639:\u0645\u0642\u0627\u0644\u0647 \u0627\u0635\u0644\u06cc \u06cc\u0627\u0646 \u06af\u0648\u062f\u0641\u0644\u0648\u0645\u0633\u062a\u0646\u062f\u0627\u062a TensorFlow\u0645\u062e\u0632\u0646 Github \u0645\u0646\u0645\u06cc \u062a\u0648\u0627\u0646\u06cc\u062f \u0633\u0648\u0627\u0644\u0627\u062a \u062e\u0648\u062f \u0631\u0627 \u0628\u067e\u0631\u0633\u06cc\u062f \u06cc\u0627 \u067e\u0631\u0648\u0698\u0647 \u0647\u0627\u06cc GAN \u062e\u0648\u062f \u0631\u0627 \u062f\u0631 \u0646\u0638\u0631\u0627\u062a \u0632\u06cc\u0631 \u0628\u0647 \u0627\u0634\u062a\u0631\u0627\u06a9 \u0628\u06af\u0630\u0627\u0631\u06cc\u062f!<\/p>\n

\n

\u0645\u0646 \u0628\u0627 \u0627\u06cc\u0646 \u0627\u06cc\u062f\u0647 \u06cc\u0627\u062f\u06af\u06cc\u0631\u06cc \u0645\u0627\u0634\u06cc\u0646\u06cc \u0628\u0627\u0634\u06a9\u0648\u0647 \u06a9\u0647 \u0628\u0647 \u0639\u0646\u0648\u0627\u0646 \u0634\u0628\u06a9\u0647\u200c\u0647\u0627\u06cc \u0645\u062a\u062e\u0627\u0635\u0645 \u0645\u0648\u0644\u062f (GAN) \u0634\u0646\u0627\u062e\u062a\u0647 \u0645\u06cc\u200c\u0634\u0648\u062f\u060c \u0628\u0647 \u062e\u0635\u0648\u0635 \u062f\u0631 \u062d\u0648\u0632\u0647 \u062a\u0648\u0644\u06cc\u062f \u062a\u0635\u0648\u06cc\u0631 \u0622\u0634\u0646\u0627 \u0634\u062f\u0645. \u0686\u0627\u0631\u0686\u0648\u0628 \u062f\u06cc\u06af\u0631\u06cc \u0628\u0647 \u0646\u0627\u0645 GANs \u062a\u0648\u0633\u0637 \u0627\u06cc\u0627\u0646 \u06af\u0648\u062f\u0641\u0644\u0648 \u062f\u0631 \u0633\u0627\u0644 2014 \u062a\u0648\u0633\u0639\u0647 \u06cc\u0627\u0641\u062a. \u0645\u0639\u0645\u0627\u0631\u06cc \u0632\u06cc\u0631\u0628\u0646\u0627\u06cc\u06cc \u0622\u0646 \u0628\u0627 \u0627\u0633\u062a\u0641\u0627\u062f\u0647 \u0627\u0632 \u0631\u0642\u0627\u0628\u062a \u062f\u0648 \u0634\u0628\u06a9\u0647 \u0639\u0635\u0628\u06cc \u0633\u0627\u062e\u062a\u0647 \u0634\u062f\u0647 \u0627\u0633\u062a. \u0628\u0627 \u062a\u0648\u062c\u0647 \u0628\u0647 \u0648\u0633\u0639\u062a \u0627\u06cc\u0646 \u0648\u0628\u0644\u0627\u06af\u060c \u0627\u062c\u0627\u0632\u0647 \u062f\u0647\u06cc\u062f \u0627\u0628\u062a\u062f\u0627 \u0645\u0639\u0631\u0641\u06cc \u06a9\u0646\u0645 GAN \u0686\u06cc\u0633\u062a\u060c \u0648 \u0633\u067e\u0633 \u0628\u0647 \u0634\u0645\u0627 \u0628\u06af\u0648\u06cc\u0645 \u06a9\u0647 \u062f\u0631 \u0627\u06cc\u0646 \u0648\u0628\u0644\u0627\u06af \u0627\u0632 \u062c\u0645\u0644\u0647 \u06a9\u062f \u0645\u0648\u062c\u0648\u062f \u062f\u0631 TensorFlow \u062f\u0631 \u0645\u0648\u0631\u062f \u0646\u062d\u0648\u0647 \u0622\u0645\u0648\u0632\u0634 \u06cc\u06a9 GAN \u0633\u0627\u062f\u0647\u060c \u0686\u0647 \u06a9\u0627\u0631 \u062e\u0648\u0627\u0647\u0645 \u06a9\u0631\u062f.<\/p>\n

<\/p>\n

GAN \u0686\u06cc\u0633\u062a\u061f<\/strong>
\u062f\u0631 \u0647\u0633\u062a\u0647 \u062e\u0648\u062f\u060c \u06cc\u06a9 GAN \u0627\u0632 \u062f\u0648 \u0634\u0628\u06a9\u0647 \u0639\u0635\u0628\u06cc \u062a\u0634\u06a9\u06cc\u0644 \u0634\u062f\u0647 \u0627\u0633\u062a: \u0627\u0644\u0628\u062a\u0647\u060c \u0645\u0648\u0644\u062f \u062f\u0627\u062f\u0647 \u0647\u0627\u06cc \u062c\u0639\u0644\u06cc \u0648\u062c\u0648\u062f \u062f\u0627\u0631\u062f\u060c \u0648 \u062a\u0645\u0627\u06cc\u0632 \u062f\u0647\u0646\u062f\u0647 \u0627\u06cc \u06a9\u0647 \u06cc\u0627\u062f \u0645\u06cc \u06af\u06cc\u0631\u062f \u0686\u06af\u0648\u0646\u0647 \u0628\u06cc\u0646 \u0686\u06cc\u0632\u0647\u0627\u06cc \u062c\u0639\u0644\u06cc \u0648 \u0648\u0627\u0642\u0639\u06cc \u062a\u0645\u0627\u06cc\u0632 \u0642\u0627\u0626\u0644 \u0634\u0648\u062f. <\/p>\n