├── .gitignore
├── CHANGELOG.md
├── LICENSE
├── README.md
├── mnist_ebgan_generate.py
├── mnist_ebgan_train.py
├── model.py
└── png
    ├── sample.png
    └── sample_with_pt.png


/.gitignore:
--------------------------------------------------------------------------------
 1 | /asset
 2 | # Byte-compiled / optimized / DLL files
 3 | __pycache__/
 4 | *.py[cod]
 5 | *$py.class
 6 | 
 7 | # C extensions
 8 | *.so
 9 | 
10 | # Distribution / packaging
11 | .Python
12 | env/
13 | build/
14 | develop-eggs/
15 | dist/
16 | downloads/
17 | eggs/
18 | .eggs/
19 | lib/
20 | lib64/
21 | parts/
22 | sdist/
23 | var/
24 | *.egg-info/
25 | .installed.cfg
26 | *.egg
27 | 
28 | # PyInstaller
29 | #  Usually these files are written by a python script from a template
30 | #  before PyInstaller builds the exe, so as to inject date/other infos into it.
31 | *.manifest
32 | *.spec
33 | 
34 | # Installer logs
35 | pip-log.txt
36 | pip-delete-this-directory.txt
37 | 
38 | # Unit test / coverage reports
39 | htmlcov/
40 | .tox/
41 | .coverage
42 | .coverage.*
43 | .cache
44 | nosetests.xml
45 | coverage.xml
46 | *,cover
47 | .hypothesis/
48 | 
49 | # Translations
50 | *.mo
51 | *.pot
52 | 
53 | # Django stuff:
54 | *.log
55 | local_settings.py
56 | 
57 | # Flask stuff:
58 | instance/
59 | .webassets-cache
60 | 
61 | # Scrapy stuff:
62 | .scrapy
63 | 
64 | # Sphinx documentation
65 | docs/_build/
66 | 
67 | # PyBuilder
68 | target/
69 | 
70 | # IPython Notebook
71 | .ipynb_checkpoints
72 | 
73 | # pyenv
74 | .python-version
75 | 
76 | # celery beat schedule file
77 | celerybeat-schedule
78 | 
79 | # dotenv
80 | .env
81 | 
82 | # virtualenv
83 | venv/
84 | ENV/
85 | 
86 | # Spyder project settings
87 | .spyderproject
88 | 
89 | # Rope project settings
90 | .ropeproject
91 | 


--------------------------------------------------------------------------------
/CHANGELOG.md:
--------------------------------------------------------------------------------
 1 | ## 0.0.0.2 ( 2017-04-04 )
 2 | 
 3 | Features :
 4 | 
 5 |    
 6 | Refactored :
 7 | 
 8 |     - adapted to tensorflow 1.0.0
 9 |     - split modeling stub from train.py to model.py
10 |      


--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
 1 | MIT License
 2 | 
 3 | Copyright (c) 2016 Namju Kim
 4 | 
 5 | Permission is hereby granted, free of charge, to any person obtaining a copy
 6 | of this software and associated documentation files (the "Software"), to deal
 7 | in the Software without restriction, including without limitation the rights
 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 9 | copies of the Software, and to permit persons to whom the Software is
10 | furnished to do so, subject to the following conditions:
11 | 
12 | The above copyright notice and this permission notice shall be included in all
13 | copies or substantial portions of the Software.
14 | 
15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 | SOFTWARE.
22 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # EBGAN
 2 | A tensorflow implementation of Junbo et al's Energy-based generative adversarial network ( EBGAN ) paper. 
 3 | ( See : [https://arxiv.org/pdf/1609.03126v2.pdf](https://arxiv.org/pdf/1609.03126v2.pdf) )
 4 | My implementation is somewhat different from original papers, for example I've used convolution layers
 5 | in both generator and discriminator instead of fully connected layers.
 6 | I think this isn't important and will not make a big difference in the final result.
 7 | 
 8 | ## Version
 9 | 
10 | Current Version : __***0.0.0.2***__
11 | 
12 | ## Dependencies ( VERSION MUST BE MATCHED EXACTLY! )
13 | 
14 | 1. tensorflow == 1.0.0 
15 | 1. sugartensor == 1.0.0.2
16 | 
17 | ## Training the network
18 | 
19 | Execute
20 | <pre><code>
21 | python mnist_ebgan_train.py
22 | </code></pre>
23 | to train the network. You can see the result ckpt files and log files in the 'asset/train' directory.
24 | Launch tensorboard --logdir asset/train/log to monitor training process.
25 | 
26 | 
27 | ## Generating image
28 |  
29 | Execute
30 | <pre><code>
31 | python mnist_ebgan_generate.py
32 | </code></pre>
33 | to generate sample image.  The 'sample.png' file will be generated in the 'asset/train' directory.
34 | 
35 | ## Generated image sample
36 | 
37 | This image was generated by EBGAN network.
38 | <p align="center">
39 |   <img src="https://raw.githubusercontent.com/buriburisuri/ebgan/master/png/sample.png" width="1024"/>
40 | </p>  
41 | 
42 | ## Other resources
43 | 
44 | 1. [Original GAN tensorflow implementation](https://github.com/buriburisuri/sugartensor/blob/master/sugartensor/example/mnist_gan.py)
45 | 1. [InfoGAN tensorflow implementation](https://github.com/buriburisuri/sugartensor/blob/master/sugartensor/example/mnist_info_gan.py)
46 | 1. [Supervised InfoGAN tensorflow implementation](https://github.com/buriburisuri/supervised_infogan)
47 | 
48 | # Authors
49 | Namju Kim (buriburisuri@gmail.com) at Jamonglabs Co., Ltd.


--------------------------------------------------------------------------------
/mnist_ebgan_generate.py:
--------------------------------------------------------------------------------
 1 | import sugartensor as tf
 2 | import matplotlib
 3 | matplotlib.use('Agg')
 4 | import matplotlib.pyplot as plt
 5 | from model import *
 6 | 
 7 | 
 8 | __author__ = 'namju.kim@kakaobrain.com'
 9 | 
10 | 
11 | # set log level to debug
12 | tf.sg_verbosity(10)
13 | 
14 | #
15 | # hyper parameters
16 | #
17 | 
18 | batch_size = 100
19 | 
20 | 
21 | # random uniform seed
22 | z = tf.random_uniform((batch_size, z_dim))
23 | 
24 | # generator
25 | gen = generator(z)
26 | 
27 | #
28 | # draw samples
29 | #
30 | 
31 | with tf.Session() as sess:
32 | 
33 |     tf.sg_init(sess)
34 | 
35 |     # restore parameters
36 |     tf.sg_restore(sess, tf.train.latest_checkpoint('asset/train'), category='generator')
37 | 
38 |     # run generator
39 |     imgs = sess.run(gen.sg_squeeze())
40 | 
41 |     # plot result
42 |     _, ax = plt.subplots(10, 10, sharex=True, sharey=True)
43 |     for i in range(10):
44 |         for j in range(10):
45 |             ax[i][j].imshow(imgs[i * 10 + j], 'gray')
46 |             ax[i][j].set_axis_off()
47 |     plt.savefig('asset/train/sample.png', dpi=600)
48 |     tf.sg_info('Sample image saved to "asset/train/sample.png"')
49 |     plt.close()
50 | 


--------------------------------------------------------------------------------
/mnist_ebgan_train.py:
--------------------------------------------------------------------------------
 1 | import sugartensor as tf
 2 | import numpy as np
 3 | from model import *
 4 | 
 5 | 
 6 | __author__ = 'namju.kim@kakaobrain.com'
 7 | 
 8 | 
 9 | # set log level to debug
10 | tf.sg_verbosity(10)
11 | 
12 | #
13 | # hyper parameters
14 | #
15 | 
16 | batch_size = 128   # batch size
17 | 
18 | #
19 | # inputs
20 | #
21 | 
22 | # MNIST input tensor ( with QueueRunner )
23 | data = tf.sg_data.Mnist(batch_size=batch_size)
24 | 
25 | # input images
26 | x = data.train.image
27 | 
28 | # random uniform seed
29 | z = tf.random_uniform((batch_size, z_dim))
30 | 
31 | #
32 | # Computational graph
33 | #
34 | 
35 | # generator
36 | gen = generator(z)
37 | 
38 | # add image summary
39 | tf.sg_summary_image(x, name='real')
40 | tf.sg_summary_image(gen, name='fake')
41 | 
42 | # discriminator
43 | disc_real = discriminator(x)
44 | disc_fake = discriminator(gen)
45 | 
46 | #
47 | # pull-away term ( PT ) regularizer
48 | #
49 | 
50 | sample = gen.sg_flatten()
51 | nom = tf.matmul(sample, tf.transpose(sample, perm=[1, 0]))
52 | denom = tf.reduce_sum(tf.square(sample), reduction_indices=[1], keep_dims=True)
53 | pt = tf.square(nom/denom)
54 | pt -= tf.diag(tf.diag_part(pt))
55 | pt = tf.reduce_sum(pt) / (batch_size * (batch_size - 1))
56 | 
57 | 
58 | #
59 | # loss & train ops
60 | #
61 | 
62 | # mean squared errors
63 | mse_real = tf.reduce_mean(tf.square(disc_real - x), reduction_indices=[1, 2, 3])
64 | mse_fake = tf.reduce_mean(tf.square(disc_fake - gen), reduction_indices=[1, 2, 3])
65 | 
66 | # discriminator loss
67 | loss_disc = mse_real + tf.maximum(margin - mse_fake, 0)
68 | # generator loss + PT regularizer
69 | loss_gen = mse_fake + pt * pt_weight
70 | 
71 | train_disc = tf.sg_optim(loss_disc, lr=0.001, category='discriminator')  # discriminator train ops
72 | train_gen = tf.sg_optim(loss_gen, lr=0.001, category='generator')  # generator train ops
73 | 
74 | # add summary
75 | tf.sg_summary_loss(loss_disc, name='disc')
76 | tf.sg_summary_loss(loss_gen, name='gen')
77 | 
78 | 
79 | #
80 | # training
81 | #
82 | 
83 | # def alternate training func
84 | @tf.sg_train_func
85 | def alt_train(sess, opt):
86 |     l_disc = sess.run([loss_disc, train_disc])[0]  # training discriminator
87 |     l_gen = sess.run([loss_gen, train_gen])[0]     # training generator
88 |     return np.mean(l_disc) + np.mean(l_gen)
89 | 
90 | # do training
91 | alt_train(log_interval=10, max_ep=30, ep_size=data.train.num_batch)
92 | 
93 | 


--------------------------------------------------------------------------------
/model.py:
--------------------------------------------------------------------------------
 1 | import sugartensor as tf
 2 | 
 3 | #
 4 | # hyper parameters
 5 | #
 6 | 
 7 | z_dim = 50        # noise dimension
 8 | margin = 1        # max-margin for hinge loss
 9 | pt_weight = 0.1   # PT regularizer's weight
10 | 
11 | 
12 | #
13 | # create generator
14 | #
15 | 
16 | def generator(x):
17 | 
18 |     reuse = len([t for t in tf.global_variables() if t.name.startswith('generator')]) > 0
19 |     with tf.sg_context(name='generator', size=4, stride=2, act='leaky_relu', bn=True, reuse=reuse):
20 | 
21 |         # generator network
22 |         res = (x.sg_dense(dim=1024, name='fc_1')
23 |                .sg_dense(dim=7*7*128, name='fc_2')
24 |                .sg_reshape(shape=(-1, 7, 7, 128))
25 |                .sg_upconv(dim=64, name='conv_1')
26 |                .sg_upconv(dim=1, act='sigmoid', bn=False, name='conv_2'))
27 |     return res
28 | 
29 | 
30 | #
31 | # create discriminator
32 | #
33 | 
34 | def discriminator(x):
35 | 
36 |     reuse = len([t for t in tf.global_variables() if t.name.startswith('discriminator')]) > 0
37 |     with tf.sg_context(name='discriminator', size=4, stride=2, act='leaky_relu', bn=True, reuse=reuse):
38 |         res = (x.sg_conv(dim=64, name='conv_1')
39 |                 .sg_conv(dim=128, name='conv_2')
40 |                 .sg_upconv(dim=64, name='conv_3')
41 |                 .sg_upconv(dim=1, act='linear', name='conv_4'))
42 | 
43 |     return res
44 | 


--------------------------------------------------------------------------------
/png/sample.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/buriburisuri/ebgan/370025905a29ccb3cad22ffdf80220dfe63f9088/png/sample.png


--------------------------------------------------------------------------------
/png/sample_with_pt.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/buriburisuri/ebgan/370025905a29ccb3cad22ffdf80220dfe63f9088/png/sample_with_pt.png


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