Fixed merge conflicts

discriminator.py

@@ -38,7 +38,7 @@ def denseW(input, output):
     return tf.Variable(tf.random_normal([input, output]))
 
 
-def conv_net(x):
+def conv_weights():
     weights = {
         'c1': filters(4, 64),
         'c2': filters(64, 128),
@@ -58,7 +58,11 @@ def conv_net(x):
         'c6': bias(512),
         'c7': bias(1),
     }
+    return (weights, biases)
 
+
+def conv_net(x, vars):
+    weights, biases = vars
     x = tf.reshape(x, shape=[-1, 128, 128, 4])
 
     c1 = conv(x, weights['c1'], biases['c1'], strides=2)
@@ -76,7 +80,9 @@ def conv_net(x):
     c7 = conv(c6, weights['c7'], biases['c7'], strides=2)
 
     res = tf.reshape(c7, [-1, 1])
-    return tf.nn.sigmoid(res)
+    vars = list(weights.values())
+    vars.extend(biases.values())
+    return (tf.nn.sigmoid(res), vars)
 
 
 def test_conv_net():

net.py

@@ -4,7 +4,22 @@
 import matplotlib.pyplot as plt
 
 from generator import u_net
-from discriminator import conv_net
+from discriminator import conv_net, conv_weights
+
+minib_size = 128
+X_dim = 128*128
+y_dim = 128*128
+Z_dim = 100
+
+X = tf.placeholder(tf.float32, shape=[None, X_dim])
+y = tf.placeholder(tf.float32, shape=[None, y_dim])
+Z = tf.placeholder(tf.float32, shape=[None, Z_dim])
+
+""" Discriminator """
+D_theta = []
+
+""" Generator """
+G_theta = []
 
 epochs = 10
 minib_size = 14
@@ -19,11 +34,11 @@
 def generator(x):
     return u_net(x)
 
-def discriminator(x, g):
+def discriminator(x, g, W, b):
     x = tf.reshape(x, [-1, 128, 128, 1])
     g = tf.reshape(g, [-1, 128, 128, 3])
     y = tf.concat([x, g], 3)
-    return conv_net(y)
+    return conv_net(y, (W, b))
 
 def next_data_batch(minibatch_size):
     pass
@@ -65,15 +80,20 @@ def next_data_batch(minibatch_size):
 
 # --> Add conditional stuff
 G_sample = generator(X_sketch) # add conditional parameter
-D_real = discriminator(X_ground_truth, X_sketch)
-D_fake = discriminator(X_ground_truth, G_sample)
+
+D_W, D_b = conv_weights()
+D_theta.extend(D_W.values())
+D_theta.extend(D_b.values())
+
+D_real = discriminator(X_ground_truth, X_sketch, (D_W, D_b))
+D_fake = discriminator(X_ground_truth, G_sample, (D_W, D_b))
 
 D_loss = -tf.reduce_mean(tf.log(D_real) + tf.log(1. - D_fake))
 G_loss = -tf.reduce_mean(tf.log(D_fake)) + tf.reduce_mean(X_sketch - D_fake)
 
 # Apply an optimizer here to minimize the above loss functions
-D_solver = tf.train.AdamOptimizer().minimize(D_loss) # --> add var_list
-G_solver = tf.train.AdamOptimizer().minimize(G_loss) # --> add var_list
+D_solver = tf.train.AdamOptimizer().minimize(D_loss, var_list = D_theta)
+G_solver = tf.train.AdamOptimizer().minimize(G_loss, var_list = G_theta)
 
 theta_D = [] ### FILL THIS IN
 theta_G = [] ### FILL THIS IN
@@ -97,9 +117,9 @@ def next_data_batch(minibatch_size):
         X_ground_truth = 0 ## Figure out how to get sketches 
 
 	_, D_loss_curr = sess.run([D_solver, D_loss],
-				  feed_dict={X_sketch: X_sketches, X_ground_truth = X_true})
+                feed_dict={X_sketch: X_sketches, X_ground_truth: X_true})
 	_, G_loss_curr = sess.run([G_solver, G_loss],
-				  feed_dict={X_ground_truth = X_true})
+                feed_dict={X_ground_truth: X_true})
 
     # Stops background threads
     coord.request_stop()