編集履歴

質問編集履歴

修正

2017/08/20 01:34

投稿

trafalbad

スコア303

test CHANGED Viewed

File without changes

test CHANGED Viewed

@@ -1,43 +1,453 @@
-下記のようにclassがあります
+下記のようにtrain.pyがあります
+その上で下記コードを実行するとエラーが出ます。
+原理的には合ってるはずなのですが、何故下記エラーが出てしますのでしょうか？
+修正させていただきました。回答よろし願いします。
 ```
+import sys
-class staff:
+import pickle
+import numpy as np
-    def __init__(self,sess):
+import tensorflow as tf
+import os
+import matplotlib.pyplot as plt
-        self.sess = sees
+import scipy.misc
-        self.build()
+import time
+from train import DCGAN
+from layer import show_all_variables
-    def build():
+def unpickle(file):
-        self.G = self.generator(self.z)
+    fp = open(file, 'rb')
+    if sys.version_info.major == 2:
-    def generator(self, z)
+        data = pickle.load(fp)
+    elif sys.version_info.major == 3:
+        data = pickle.load(fp, encoding='latin-1')
+    fp.close()
-        return z
+    return data
+test= unpickle("train_image.pickle")
+X_image=np.array(test)/127.5 - 1
+run_config = tf.ConfigProto()
+run_config.gpu_options.allow_growth=True
+with tf.Session(config=run_config) as sess:
+    dcgan = DCGAN(sess, X_image=X_image, epochs=10, step=20)
+    show_all_variables()
+    dcgan.train()
+---------------------------------------------------------------------------
+NameError                                 Traceback (most recent call last)
+<ipython-input-16-fb64556ff7b2> in <module>()
+      3
+      4 with tf.Session(config=run_config) as sess:
+----> 5     dcgan = DCGAN(sess, X_image=X_image, epochs=10, step=20)
+      6     show_all_variables()
+      7     dcgan.train()
+~/DCGAN-tensorflow/train.py in __init__(self, sess, X_image, epochs, step)
+     27         self.step = step
+     28
+---> 29         self.build_model()
+     30
+     31
+~/DCGAN-tensorflow/train.py in build_model(self)
+     35         batch_size=64
+     36         with tf.variable_scope("discriminator") as scope:
+---> 37             h0 = lrelu(conv2d(image, 64, name='d_h0_conv'))
+     38             h1 = lrelu(batch_norm(conv2d(h0, 128, name='d_h1_conv'),'d_bn1'))
+     39             h2 = lrelu(batch_norm(conv2d(h1, 256, name='d_h2_conv'),'d_bn2'))
+NameError: name 'generator' is not defined
 ```
-staffクラス内の```generator```関数を同クラス内の```build```関数内で上記のように使うにはどのようにしたら良いのでしょうか？
-上記の書き方だと
-```NameError: name 'generator' is not defined```
-となってしまうのですが
+```
+train.py
+# coding: utf-8
+# In[1]:
+import sys
+import pickle
+import numpy as np
+import tensorflow as tf
+import os
+import matplotlib.pyplot as plt
+import scipy.misc
+import time
+from layer import *
+# In[2]:
+class DCGAN(object):
+    def __init__(self, sess, X_image, epochs, step):
+        self.sess = sess
+        self.X_image=X_image
+        self.epochs = epochs
+        self.step = step
+        self.build_model()
+    def discriminator(self, image):
+        batch_size=64
+        with tf.variable_scope("discriminator") as scope:
+            h0 = lrelu(conv2d(image, 64, name='d_h0_conv'))
+            h1 = lrelu(batch_norm(conv2d(h0, 128, name='d_h1_conv'),'d_bn1'))
+            h2 = lrelu(batch_norm(conv2d(h1, 256, name='d_h2_conv'),'d_bn2'))
+            h3 = lrelu(batch_norm(conv2d(h2, 512, name='d_h3_conv'),'d_bn3'))  # shape=(batch_size, 64, 64, 3)
+            h4 = linear_d(tf.reshape(h3, [batch_size, -1]),2,'d_h4_lin')
+            return h4
+    # shape=(batch_size, 64, 64, 3)
+    def generator(self, z_):
+        batch_size=64
+        with tf.variable_scope("generator") as scope:
+            # project `z` and reshape
+            z, h0_w, h0_b = linear(z_, 64*8*4*4, 'g_h0_lin',with_w=True)
+            h0 = tf.nn.relu(batch_norm(tf.reshape(z, [-1, 4, 4, 64*8]), 'g_bn0'))
+            h1, h1_w, h1_b = deconv2d(h0, [batch_size, 8, 8, 64*4], name='g_h1', with_w=True)
+            h1 = tf.nn.relu(batch_norm(h1, 'g_bn1'))
+            h2, h2_w, h2_b = deconv2d(h1, [batch_size, 16, 16, 64*2], name='g_h2', with_w=True)
+            h2 = tf.nn.relu(batch_norm(h2, 'g_bn2'))
+            h3, h3_w, h3_b = deconv2d(h2, [batch_size, 32, 32, 64*1], name='g_h3', with_w=True)
+            h3 = tf.nn.relu(batch_norm(h3, 'g_bn3'))
+            h4, h4_w, h4_b = deconv2d(h3, [batch_size, 64, 64, 3], name='g_h4', with_w=True)
+            return tf.nn.tanh(h4)  #shape=(batch_size, 64, 64, 3)
+    def sampler(self, z_):# shape=(batch_size, 64, 64, 3)
+        batch_size=64
+        with tf.variable_scope("generator") as scope:
+            # project `z` and reshape
+            z= linear(z_, 64*8*4*4,'g_h0_lin')
+            h0 = tf.nn.relu(batch_norm(tf.reshape(z, [-1, 4, 4, 64*8]),'g_bn0',train=False))
+            h1 = deconv2d(h0, [batch_size, 8, 8, 64*4], name='g_h1')
+            h1 = tf.nn.relu(batch_norm(h1,'g_bn1',train=False))
+            h2 = deconv2d(h1, [batch_size, 16, 16, 64*2], name='g_h2')
+            h2 = tf.nn.relu(batch_norm(h2,'g_bn2',train=False))
+            h3 = deconv2d(h2, [batch_size, 32, 32, 64*1], name='g_h3')
+            h3 = tf.nn.relu(batch_norm(h3,'g_bn3',train=False))
+            h4 = deconv2d(h3, [batch_size, 64, 64, 3], name='g_h4')
+            return tf.nn.tanh(h4)  #shape=(batch_size, 64, 64, 3)
+    def build_model(self):
+        self.z = tf.placeholder(tf.float32, [64, 100])
+        self.image = tf.placeholder(tf.float32, [64, 64, 64, 3])
+        self.G=self.generator(self.z)  #G(z)
+        self.D_logits = self.discriminator(self.image) #D(x)
+        self.sampler = self.sampler(self.z)
+        self.D_logits_ = self.discriminator(self.G)   #D(G(z))
+        batch_label=64
+        self.d_loss_real = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=self.D_logits, labels=tf.ones([batch_label], dtype=tf.int64)))
+        self.d_loss_fake = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=self.D_logits_, labels=tf.zeros([batch_label], dtype=tf.int64)))
+        self.g_loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=self.D_logits_, labels=tf.ones([batch_label], dtype=tf.int64)))
+        self.d_loss = self.d_loss_real + self.d_loss_fake
+        self.d_vars = [var for var in tf.trainable_variables() if 'd_' in var.name]
+        self.g_vars = [var for var in tf.trainable_variables() if 'g_' in var.name]
+        self.saver=tf.train.Saver()
+    def train(self):
+        g_optim = tf.train.AdamOptimizer(learning_rate=0.0002, beta1=0.5).minimize(self.g_loss, var_list=self.g_vars)
+        d_optim = tf.train.AdamOptimizer(learning_rate=0.0002, beta1=0.5).minimize(self.d_loss, var_list=self.d_vars)
+        tf.global_variables_initializer().run()
+        sample_z = np.random.uniform(-1, 1, size=(64, 100))
+        batch_z = np.random.uniform(-1, 1, [64, 100])
+        sample_files = self.X_image[0:64]
+        sample = [sample_file for sample_file in sample_files]
+        sample_images = np.array(sample).astype(np.float32)
+        counter=1
+        start_time=time.time()
+        for epoch in range(self.epochs):
+            batch_idxs= min (len(self.X_image), np.inf) // 64
+            for idx in range (0, batch_idxs):
+                bacth_files= self.X_image[idx*64:(idx+1)*64]
+                batch = [batch_file for batch_file in bacth_files]
+                batch_images = np.array(batch).astype(np.float32)
+                self.sess.run(d_optim, feed_dict = {self.z: batch_z, self.image: batch_images})
+                self.sess.run(g_optim, feed_dict = {self.z: batch_z})
+                # Run g_optim twice to realize loss value
+                self.sess.run(g_optim, feed_dict = {self.z: batch_z})
+                errD_fake = self.d_loss_fake.eval({self.z: batch_z })
+                errD_real = self.d_loss_real.eval({self.image: batch_images})
+                errG = self.g_loss.eval({self.z: batch_z})
+                counter += 1
+                print("Epoch: [%2d] [%4d/%4d] time:%4.4f, d_loss: %.8f, g_loss: %.8f" % (epoch, idx, batch_idxs,
+                                                                                         time.time()-start_time, errD_fake+errD_real, errG))
+                if np.mod(counter, self.step)==1:
+                    samples, d_loss_sample, g_loss_sample = sess.run([self.sampler, self.d_loss, self.g_loss],
+                                               feed_dict={self.z: sample_z, self.image: sample_images})
+                    print("[Sample] d_loss:%.8f, g_loss:%.8f" % (d_loss_sample, g_loss_sample))
+                    col=8
+                    rows=[]
+                    for i in range(8):
+                        rows.append(np.hstack(samples[col * i + 0:col * i + col]))
+                    vnari=np.vstack(rows)
+                    plt.imshow(vnari)
+                    plt.show()
+```