Fix reuse variables issue and some updates for new TF

README.md

@@ -17,5 +17,6 @@ Set `train = False` to visualize how adjusting the initial noise affects image g
 
 Set `cifar = True` to train on CIFAR. This sets the network to train on 32x32 images instead of 64x64, and reads from the CIFAR binaries rather than from JPEGS.
 
+To download the CIFAR dataset, just run `wget https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz` and `tar -xzvf cifar-10-python.tar.gz`
 
 Utils and network structure from [DCGAN-Tensorflow](https://github.com/carpedm20/DCGAN-tensorflow).

main.py

@@ -8,25 +8,25 @@
 from scipy.misc import imsave as ims
 from random import randint
 
-cifar = False
+cifar = True
 
 def discriminator(image, reuse=False):
-    if reuse:
-            tf.get_variable_scope().reuse_variables()
 
     if cifar:
-        h0 = lrelu(conv2d(image, 3, df_dim, name='d_h0_conv')) #16x16x64
-        h1 = lrelu(d_bn1(conv2d(h0, df_dim, df_dim*2, name='d_h1_conv'))) #8x8x128
-        h2 = lrelu(d_bn2(conv2d(h1, df_dim*2, df_dim*4, name='d_h2_conv'))) #4x4x256
-        h4 = dense(tf.reshape(h2, [batchsize, -1]), 4*4*df_dim*4, 1, scope='d_h3_lin')
-        return tf.nn.sigmoid(h4), h4
+        with tf.variable_scope(tf.get_variable_scope(), reuse=reuse) as scope:
+            h0 = lrelu(conv2d(image, 3, df_dim, name='d_h0_conv')) #16x16x64
+            h1 = lrelu(d_bn1(conv2d(h0, df_dim, df_dim*2, name='d_h1_conv'))) #8x8x128
+            h2 = lrelu(d_bn2(conv2d(h1, df_dim*2, df_dim*4, name='d_h2_conv'))) #4x4x256
+            h4 = dense(tf.reshape(h2, [batchsize, -1]), 4*4*df_dim*4, 1, scope='d_h3_lin')
+            return tf.nn.sigmoid(h4), h4
     else:
-        h0 = lrelu(conv2d(image, 3, df_dim, name='d_h0_conv'))
-        h1 = lrelu(d_bn1(conv2d(h0, 64, df_dim*2, name='d_h1_conv')))
-        h2 = lrelu(d_bn2(conv2d(h1, 128, df_dim*4, name='d_h2_conv')))
-        h3 = lrelu(d_bn3(conv2d(h2, 256, df_dim*8, name='d_h3_conv')))
-        h4 = dense(tf.reshape(h3, [batchsize, -1]), 4*4*512, 1, scope='d_h3_lin')
-        return tf.nn.sigmoid(h4), h4
+        with tf.variable_scope(tf.get_variable_scope(), reuse=reuse) as scope:
+            h0 = lrelu(conv2d(image, 3, df_dim, name='d_h0_conv'))
+            h1 = lrelu(d_bn1(conv2d(h0, 64, df_dim*2, name='d_h1_conv')))
+            h2 = lrelu(d_bn2(conv2d(h1, 128, df_dim*4, name='d_h2_conv')))
+            h3 = lrelu(d_bn3(conv2d(h2, 256, df_dim*8, name='d_h3_conv')))
+            h4 = dense(tf.reshape(h3, [batchsize, -1]), 4*4*512, 1, scope='d_h3_lin')
+            return tf.nn.sigmoid(h4), h4
 
 def generator(z):
     if cifar:
@@ -82,10 +82,10 @@ def generator(z):
 
     D_fake_prob, D_fake_logit = discriminator(G, reuse=True)
 
-    d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(D_logit, tf.ones_like(D_logit)))
-    d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(D_fake_logit, tf.zeros_like(D_fake_logit)))
+    d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_logit, labels=tf.ones_like(D_logit)))
+    d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_fake_logit, labels=tf.zeros_like(D_fake_logit)))
 
-    gloss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(D_fake_logit, tf.ones_like(D_fake_logit)))
+    gloss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_fake_logit, labels=tf.ones_like(D_fake_logit)))
     dloss = d_loss_real + d_loss_fake
 
     t_vars = tf.trainable_variables()
@@ -102,7 +102,7 @@ def generator(z):
 
     d_optim = tf.train.AdamOptimizer(learningrate, beta1=beta1).minimize(dloss, var_list=d_vars)
     g_optim = tf.train.AdamOptimizer(learningrate, beta1=beta1).minimize(gloss, var_list=g_vars)
-    tf.initialize_all_variables().run()
+    tf.global_variables_initializer().run()
 
     saver = tf.train.Saver(max_to_keep=10)
 
@@ -111,10 +111,11 @@ def generator(z):
 
     display_z = np.random.uniform(-1, 1, [batchsize, z_dim]).astype(np.float32)
 
-    realfiles = data[0:64]
-    realim = [get_image(batch_file, [64,64,3], is_crop=False) for batch_file in realfiles]
-    real_img = np.array(realim).astype(np.float32)
-    ims("results/imagenet/real.jpg",merge(real_img,[8,8]))
+    if not cifar:
+        realfiles = data[0:64]
+        realim = [get_image(batch_file, [64,64,3], is_crop=False) for batch_file in realfiles]
+        real_img = np.array(realim).astype(np.float32)
+        ims("results/imagenet/real.jpg",merge(real_img,[8,8]))
 
     train = True
     if train:

ops.py

@@ -15,7 +15,7 @@ def __call__(self, x, train=True):
         shape = x.get_shape().as_list()
 
         if train:
-            with tf.variable_scope(self.name) as scope:
+            with tf.variable_scope(self.name, reuse=tf.AUTO_REUSE) as scope:
                 self.beta = tf.get_variable("beta", [shape[-1]],
                                     initializer=tf.constant_initializer(0.))
                 self.gamma = tf.get_variable("gamma", [shape[-1]],