├── README.md
└── gan.py


/README.md:
--------------------------------------------------------------------------------
1 | # Introduction-to-GANs-with-Python-and-TensorFlow


--------------------------------------------------------------------------------
/gan.py:
--------------------------------------------------------------------------------
  1 | 
  2 | # coding: utf-8
  3 | 
  4 | # In[1]:
  5 | 
  6 | import tensorflow as tf
  7 | from tensorflow.examples.tutorials.mnist import input_data
  8 | import numpy as np
  9 | import matplotlib.pyplot as plt
 10 | import matplotlib.gridspec as gridspec
 11 | import os
 12 | 
 13 | 
 14 | # In[2]:
 15 | 
 16 | # Sample z from uniform distribution
 17 | def sample_Z(m, n):
 18 |     return np.random.uniform(-1., 1., size=[m, n])
 19 | 
 20 | 
 21 | def plot(samples):
 22 |     fig = plt.figure(figsize=(4, 4))
 23 |     gs = gridspec.GridSpec(4, 4)
 24 |     gs.update(wspace=0.05, hspace=0.05)
 25 | 
 26 |     for i, sample in enumerate(samples):
 27 |         ax = plt.subplot(gs[i])
 28 |         plt.axis('off')
 29 |         ax.set_xticklabels([])
 30 |         ax.set_yticklabels([])
 31 |         ax.set_aspect('equal')
 32 |         plt.imshow(sample.reshape(28, 28), cmap='Greys_r')
 33 | 
 34 |     return fig
 35 | 
 36 | 
 37 | # In[3]:
 38 | 
 39 | # Declare inputs and parameters to the model
 40 | 
 41 | # Input image, foe discrminator model.
 42 | X = tf.placeholder(tf.float32, shape=[None, 784])
 43 | 
 44 | # Input noise for generator.
 45 | Z = tf.placeholder(tf.float32, shape=[None, 100])
 46 | 
 47 | 
 48 | # In[8]:
 49 | 
 50 | def generator(z):
 51 |     
 52 |     with tf.variable_scope("generator", reuse=tf.AUTO_REUSE):
 53 |         
 54 |         x = tf.layers.dense(z, 128, activation = tf.nn.relu)
 55 |         x = tf.layers.dense(z, 784)
 56 |         x = tf.nn.sigmoid(x)
 57 | 
 58 |     return x
 59 | 
 60 | 
 61 | def discriminator(x):
 62 |     with tf.variable_scope("discrminator", reuse=tf.AUTO_REUSE):
 63 |         x = tf.layers.dense(x, 128, activation = tf.nn.relu)
 64 |         x = tf.layers.dense(x, 1)
 65 |         x = tf.nn.sigmoid(x)
 66 | 
 67 |     return x
 68 | 
 69 | 
 70 | # In[10]:
 71 | 
 72 | G_sample = generator(Z)
 73 | D_real = discriminator(X)
 74 | D_fake = discriminator(G_sample)
 75 | 
 76 | D_loss = -tf.reduce_mean(tf.log(D_real) + tf.log(1. - D_fake))
 77 | G_loss = -tf.reduce_mean(tf.log(D_fake))
 78 | 
 79 | # Optimizers
 80 | disc_vars = [var for var in tf.trainable_variables() if var.name.startswith("disc")]
 81 | gen_vars =  [var for var in tf.trainable_variables() if var.name.startswith("gen")]
 82 | 
 83 | D_solver = tf.train.AdamOptimizer().minimize(D_loss, var_list=disc_vars)
 84 | G_solver = tf.train.AdamOptimizer().minimize(G_loss, var_list=gen_vars)
 85 | 
 86 | 
 87 | # In[11]:
 88 | 
 89 | # Batch size
 90 | mb_size = 128
 91 | 
 92 | # Dimention of input noise
 93 | Z_dim = 100
 94 | 
 95 | mnist = input_data.read_data_sets('../../MNIST_data', one_hot=True)
 96 | 
 97 | sess = tf.Session()
 98 | sess.run(tf.global_variables_initializer())
 99 | 
100 | if not os.path.exists('out2/'):
101 |     os.makedirs('out2/')
102 | 
103 | i = 0
104 | 
105 | for it in range(1000000):
106 |     
107 |     # Save generated images every 1000 iterations.
108 |     if it % 1000 == 0:
109 |         samples = sess.run(G_sample, feed_dict={Z: sample_Z(16, Z_dim)})
110 | 
111 |         fig = plot(samples)
112 |         plt.savefig('out2/{}.png'.format(str(i).zfill(3)), bbox_inches='tight')
113 |         i += 1
114 |         plt.close(fig)
115 |         
116 |         
117 |     # Get next batch of images. Each batch has mb_size samples.
118 |     X_mb, _ = mnist.train.next_batch(mb_size)
119 |     
120 |     
121 |     # Run disciminator solver
122 |     _, D_loss_curr = sess.run([D_solver, D_loss], feed_dict={X: X_mb, Z: sample_Z(mb_size, Z_dim)})
123 |     
124 |     # Run generator solver
125 |     _, G_loss_curr = sess.run([G_solver, G_loss], feed_dict={Z: sample_Z(mb_size, Z_dim)})
126 | 
127 |     # Print loss
128 |     if it % 1000 == 0:
129 |         print('Iter: {}'.format(it))
130 |         print('D loss: {:.4}'. format(D_loss_curr))
131 | 
132 | 
133 | 
134 | 
135 | 


--------------------------------------------------------------------------------