initial commit

README.md

@@ -1,2 +1,37 @@
-# ebgan
-tensorflow implementation of Junbo et al's Energy-based generative adversarial network ( EBGAN ) paper. 
+# EBGAN
+A tensorflow implementation of Junbo et al's Energy-based generative adversarial network ( EBGAN ) paper. 
+( See : [https://arxiv.org/pdf/1609.03126v2.pdf](https://arxiv.org/pdf/1609.03126v2.pdf) )
+
+## Dependencies
+
+1. tensorflow >= rc0.10 
+1. sugartensor >= 0.0.1
+
+## Training the network
+
+Execute
+<pre><code>
+python mnist_ebgan_train.py
+</code></pre>
+to train the network. You can see the result ckpt files and log files in the 'asset/train' directory.
+Launch tensorboard --logdir asset/train/log to monitor training process.
+
+
+## Generating image.
+
+Execute
+<pre><code>
+python mnist_ebgan_generate.py
+</code></pre>
+to generate sample image.  The 'sample.png' file will be generated in the 'asset/train' directory.
+
+## Sample image.
+
+This image was generated by EBGAN network.
+<p align="center">
+  <img src="https://raw.githubusercontent.com/buriburisuri/ebgan/master/ebgan/png/sample.png" width="350"/>
+</p>  
+
+
+# Authors
+Namju Kim ([email protected]) at Jamonglabs Co., Ltd.

mnist_ebgan_generate.py

@@ -1,6 +1,5 @@
 # -*- coding: utf-8 -*-
 import sugartensor as tf
-import numpy as np
 import matplotlib.pyplot as plt
 
 # set log level to debug

mnist_ebgan_train.py

@@ -9,9 +9,10 @@
 # hyper parameters
 #
 
-batch_size = 32
-z_dim = 50
-margin = 1
+batch_size = 32   # batch size
+z_dim = 50        # noise dimension
+margin = 1        # max-margin for hinge loss
+pt_weight = 0.1  # PT regularizer's weight
 
 #
 # inputs
@@ -23,12 +24,6 @@
 # input images
 x = data.train.image
 
-# generator labels ( all ones )
-y = tf.ones(batch_size, dtype=tf.sg_floatx)
-
-# discriminator labels ( half 1s, half 0s )
-y_disc = tf.concat(0, [y, y * 0])
-
 #
 # create generator
 #
@@ -45,7 +40,6 @@
            .sg_upconv(dim=64)
            .sg_upconv(dim=1, act='sigmoid', bn=False))
 
-
 #
 # create discriminator
 #
@@ -59,17 +53,27 @@
             .sg_upconv(dim=64)
             .sg_upconv(dim=1, act='linear'))
 
+#
+# pull-away term ( PT ) regularizer
+#
+
+sample = gen.sg_flatten()
+nom = tf.matmul(sample, tf.transpose(sample, perm=[1, 0]))
+denom = tf.reduce_sum(tf.square(sample), reduction_indices=[1], keep_dims=True)
+pt = tf.square(nom/denom)
+pt -= tf.diag(tf.diag_part(pt))
+pt = tf.reduce_sum(pt) / (batch_size * (batch_size - 1))
 
 #
 # loss & train ops
 #
 
-# squared errors
-mse = tf.square(disc - xx)
-mse_real, mse_fake = mse[:batch_size, :, :, :], mse[batch_size:, :, :, :]
+# mean squared errors
+mse = tf.reduce_mean(tf.square(disc - xx), reduction_indices=[1, 2, 3])
+mse_real, mse_fake = mse[:batch_size], mse[batch_size:]
 
 loss_disc = mse_real + tf.maximum(margin - mse_fake, 0)   # discriminator loss
-loss_gen = mse_fake  # generator loss
+loss_gen = mse_fake + pt * pt_weight   # generator loss + PT regularizer
 
 train_disc = tf.sg_optim(loss_disc, lr=0.001, category='discriminator')  # discriminator train ops
 train_gen = tf.sg_optim(loss_gen, lr=0.001, category='generator')  # generator train ops