add data

.gitignore

@@ -0,0 +1,7 @@
+__pycache__
+GANs/*.pyc
+checkpoint
+logs
+data/minist
+
+results

ACGAN.py → GANs/ACGAN.py

@@ -5,8 +5,8 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
 class ACGAN(object):
     model_name = "ACGAN"     # name for checkpoint

BEGAN.py → GANs/BEGAN.py

@@ -5,8 +5,8 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
 class BEGAN(object):
     model_name = "BEGAN"     # name for checkpoint

CGAN.py → GANs/CGAN.py

@@ -5,8 +5,8 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
 class CGAN(object):
     model_name = "CGAN"     # name for checkpoint

CVAE.py → GANs/CVAE.py

@@ -5,10 +5,10 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
-import prior_factory as prior
+import GANs.prior_factory as prior
 
 class CVAE(object):
     model_name = "CVAE"     # name for checkpoint

DRAGAN.py → GANs/DRAGAN.py

@@ -5,8 +5,8 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
 class DRAGAN(object):
     model_name = "DRAGAN"     # name for checkpoint

EBGAN.py → GANs/EBGAN.py

@@ -5,8 +5,8 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
 class EBGAN(object):
     model_name = "EBGAN"     # name for checkpoint

GAN.py → GANs/GAN.py

@@ -5,8 +5,8 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
 class GAN(object):
     model_name = "GAN"     # name for checkpoint

LSGAN.py → GANs/LSGAN.py

@@ -5,8 +5,8 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
 class LSGAN(object):
     model_name = "LSGAN"     # name for checkpoint

VAE.py → GANs/VAE.py

@@ -5,10 +5,10 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
-import prior_factory as prior
+import GANs.prior_factory as prior
 
 class VAE(object):
     model_name = "VAE"     # name for checkpoint
@@ -96,7 +96,7 @@ def build_model(self):
 
         """ Loss Function """
         # encoding
-        self.mu, sigma = self.encoder(self.inputs, is_training=True, reuse=False)        
+        self.mu, sigma = self.encoder(self.inputs, is_training=True, reuse=False)
 
         # sampling by re-parameterization technique
         z = self.mu + sigma * tf.random_normal(tf.shape(self.mu), 0, 1, dtype=tf.float32)
@@ -241,7 +241,7 @@ def visualize_results(self, epoch):
                 else:
                     z_tot = np.concatenate((z_tot, z), axis=0)
                     id_tot = np.concatenate((id_tot, batch_labels), axis=0)
-
+            # in conda, py2.7 and matpltlib version not match
             save_scattered_image(z_tot, id_tot, -4, 4, name=check_folder(
                 self.result_dir + '/' + self.model_dir) + '/' + self.model_name + '_epoch%03d' % epoch + '_learned_manifold.png')
 

WGAN.py → GANs/WGAN.py

@@ -5,8 +5,8 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
 class WGAN(object):
     model_name = "WGAN"     # name for checkpoint

WGAN_GP.py → GANs/WGAN_GP.py

@@ -5,8 +5,8 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
 class WGAN_GP(object):
     model_name = "WGAN_GP"     # name for checkpoint

infoGAN.py → GANs/infoGAN.py

@@ -5,8 +5,8 @@
 import tensorflow as tf
 import numpy as np
 
-from ops import *
-from utils import *
+from GANs.ops import *
+from GANs.utils import *
 
 class infoGAN(object):
     model_name = "infoGAN"     # name for checkpoint

ops.py → GANs/ops.py

@@ -2,12 +2,12 @@
 Most codes from https://github.com/carpedm20/DCGAN-tensorflow
 """
 import math
-import numpy as np 
+import numpy as np
 import tensorflow as tf
 
 from tensorflow.python.framework import ops
 
-from utils import *
+from GANs.utils import *
 
 if "concat_v2" in dir(tf):
     def concat(tensors, axis, *args, **kwargs):

utils.py → GANs/utils.py

@@ -9,7 +9,7 @@
 import numpy as np
 from time import gmtime, strftime
 from six.moves import xrange
-import matplotlib.pyplot as plt
+#import matplotlib.pyplot as plt
 import os, gzip
 
 import tensorflow as tf
@@ -19,6 +19,7 @@ def load_mnist(dataset_name):
     data_dir = os.path.join("./data", dataset_name)
 
     def extract_data(filename, num_data, head_size, data_size):
+        print('reading gz data {}'.format(filename))
         with gzip.open(filename) as bytestream:
             bytestream.read(head_size)
             buf = bytestream.read(data_size * num_data)
@@ -123,27 +124,31 @@ def inverse_transform(images):
     return (images+1.)/2.
 
 """ Drawing Tools """
+# in conda, py2.7 and matpltlib version not match, del
+
 # borrowed from https://github.com/ykwon0407/variational_autoencoder/blob/master/variational_bayes.ipynb
-def save_scattered_image(z, id, z_range_x, z_range_y, name='scattered_image.jpg'):
-    N = 10
-    plt.figure(figsize=(8, 6))
-    plt.scatter(z[:, 0], z[:, 1], c=np.argmax(id, 1), marker='o', edgecolor='none', cmap=discrete_cmap(N, 'jet'))
-    plt.colorbar(ticks=range(N))
-    axes = plt.gca()
-    axes.set_xlim([-z_range_x, z_range_x])
-    axes.set_ylim([-z_range_y, z_range_y])
-    plt.grid(True)
-    plt.savefig(name)
-
-# borrowed from https://gist.github.com/jakevdp/91077b0cae40f8f8244a
-def discrete_cmap(N, base_cmap=None):
-    """Create an N-bin discrete colormap from the specified input map"""
-
-    # Note that if base_cmap is a string or None, you can simply do
-    #    return plt.cm.get_cmap(base_cmap, N)
-    # The following works for string, None, or a colormap instance:
-
-    base = plt.cm.get_cmap(base_cmap)
-    color_list = base(np.linspace(0, 1, N))
-    cmap_name = base.name + str(N)
-    return base.from_list(cmap_name, color_list, N)
+# def save_scattered_image(z, id, z_range_x, z_range_y, name='scattered_image.jpg'):
+#     N = 10
+#     plt.figure(figsize=(8, 6))
+#     plt.scatter(z[:, 0], z[:, 1], c=np.argmax(id, 1), marker='o', edgecolor='none', cmap=discrete_cmap(N, 'jet'))
+#     plt.colorbar(ticks=range(N))
+#     axes = plt.gca()
+#     axes.set_xlim([-z_range_x, z_range_x])
+#     axes.set_ylim([-z_range_y, z_range_y])
+#     plt.grid(True)
+#     plt.savefig(name)
+
+# # borrowed from https://gist.github.com/jakevdp/91077b0cae40f8f8244a
+# def discrete_cmap(N, base_cmap=None):
+#     """Create an N-bin discrete colormap from the specified input map"""
+
+#     # Note that if base_cmap is a string or None, you can simply do
+#     #    return plt.cm.get_cmap(base_cmap, N)
+#     # The following works for string, None, or a colormap instance:
+
+#     base = plt.cm.get_cmap(base_cmap)
+#     color_list = base(np.linspace(0, 1, N))
+#     cmap_name = base.name + str(N)
+#     return base.from_list(cmap_name, color_list, N)
+
+

README.md

@@ -1,16 +1,17 @@
 # tensorflow-generative-model-collections
 Tensorflow implementation of various GANs and VAEs.
 
+
 ## Related Repositories
 ### Pytorch version
 Pytorch version of this repository is availabel at https://github.com/znxlwm/pytorch-generative-model-collections
 
-### "Are GANs Created Equal? A Large-Scale Study" Paper 
-https://github.com/google/compare_gan is the code that was used in [the paper](https://arxiv.org/abs/1711.10337).  
-It provides IS/FID and rich experimental results for all gan-variants.  
+### "Are GANs Created Equal? A Large-Scale Study" Paper
+https://github.com/google/compare_gan is the code that was used in [the paper](https://arxiv.org/abs/1711.10337).
+It provides IS/FID and rich experimental results for all gan-variants.
 
 ## Generative Adversarial Networks (GANs)
-### Lists  
+### Lists
 
 *Name* | *Paper Link* | *Value Function*
 :---: | :---: | :--- |
@@ -23,16 +24,16 @@ It provides IS/FID and rich experimental results for all gan-variants.
 **infoGAN**| [Arxiv](https://arxiv.org/abs/1606.03657) | <img src = 'assets/equations/infoGAN.png' height = '70px'>
 **ACGAN**| [Arxiv](https://arxiv.org/abs/1610.09585) | <img src = 'assets/equations/ACGAN.png' height = '70px'>
 **EBGAN**| [Arxiv](https://arxiv.org/abs/1609.03126) | <img src = 'assets/equations/EBGAN.png' height = '70px'>
-**BEGAN**| [Arxiv](https://arxiv.org/abs/1702.08431) | <img src = 'assets/equations/BEGAN.png' height = '105px'>  
+**BEGAN**| [Arxiv](https://arxiv.org/abs/1702.08431) | <img src = 'assets/equations/BEGAN.png' height = '105px'>
 
 #### Variants of GAN structure
 <img src = 'assets/etc/GAN_structure.png' height = '600px'>
 
 ### Results for mnist
-Network architecture of generator and discriminator is the exaclty sames as in [infoGAN paper](https://arxiv.org/abs/1606.03657).  
+Network architecture of generator and discriminator is the exaclty sames as in [infoGAN paper](https://arxiv.org/abs/1606.03657).
 For fair comparison of core ideas in all gan variants, all implementations for network architecture are kept same except EBGAN and BEGAN. Small modification is made for EBGAN/BEGAN, since those adopt auto-encoder strucutre for discriminator. But I tried to keep the capacity of discirminator.
 
-The following results can be reproduced with command:  
+The following results can be reproduced with command:
 ```
 python main.py --dataset mnist --gan_type <TYPE> --epoch 25 --batch_size 64
 ```
@@ -68,10 +69,10 @@ infoGAN | <img src = 'assets/mnist_results/conditional_generation/infoGAN_epoch0
 </table>
 
 ### Results for fashion-mnist
-Comments on network architecture in mnist are also applied to here.  
+Comments on network architecture in mnist are also applied to here.
 [Fashion-mnist](https://github.com/zalandoresearch/fashion-mnist) is a recently proposed dataset consisting of a training set of 60,000 examples and a test set of 10,000 examples. Each example is a 28x28 grayscale image, associated with a label from 10 classes. (T-shirt/top, Trouser, Pullover, Dress, Coat, Sandal, Shirt, Sneaker, Bag, Ankle boot)
 
-The following results can be reproduced with command:  
+The following results can be reproduced with command:
 ```
 python main.py --dataset fashion-mnist --gan_type <TYPE> --epoch 40 --batch_size 64
 ```
@@ -98,8 +99,8 @@ CGAN | <img src = 'assets/fashion_mnist_results/conditional_generation/CGAN_epoc
 ACGAN | <img src = 'assets/fashion_mnist_results/conditional_generation/ACGAN_epoch000_test_all_classes_style_by_style.png' height = '230px'> | <img src = 'assets/fashion_mnist_results/conditional_generation/ACGAN_epoch019_test_all_classes_style_by_style.png' height = '230px'> | <img src = 'assets/fashion_mnist_results/conditional_generation/ACGAN_epoch039_test_all_classes_style_by_style.png' height = '230px'>
 infoGAN | <img src = 'assets/fashion_mnist_results/conditional_generation/infoGAN_epoch000_test_all_classes_style_by_style.png' height = '230px'> | <img src = 'assets/fashion_mnist_results/conditional_generation/infoGAN_epoch019_test_all_classes_style_by_style.png' height = '230px'> | <img src = 'assets/fashion_mnist_results/conditional_generation/infoGAN_epoch039_test_all_classes_style_by_style.png' height = '230px'>
 
-Without hyper-parameter tuning from mnist-version, ACGAN/infoGAN does not work well as compared with CGAN.  
-ACGAN tends to fall into mode-collapse.  
+Without hyper-parameter tuning from mnist-version, ACGAN/infoGAN does not work well as compared with CGAN.
+ACGAN tends to fall into mode-collapse.
 infoGAN tends to ignore noise-vector. It results in that various style within the same class can not be represented.
 
 #### InfoGAN : Manipulating two continous codes
@@ -122,15 +123,15 @@ infoGAN tends to ignore noise-vector. It results in that various style within th
 **VAE**| [Arxiv](https://arxiv.org/abs/1312.6114) | <img src = 'assets/equations/VAE.png' height = '35px'>
 **CVAE**| [Arxiv](https://arxiv.org/abs/1406.5298) | <img src = 'assets/equations/CVAE.png' height = '35px'>
 **DVAE**| [Arxiv](https://arxiv.org/abs/1511.06406) | (to be added)
-**AAE**| [Arxiv](https://arxiv.org/abs/1511.05644) | (to be added) 
+**AAE**| [Arxiv](https://arxiv.org/abs/1511.05644) | (to be added)
 
 #### Variants of VAE structure
 <img src = 'assets/etc/VAE_structure.png' height = '280px'>
 
 ### Results for mnist
 Network architecture of decoder(generator) and encoder(discriminator) is the exaclty sames as in [infoGAN paper](https://arxiv.org/abs/1606.0365). The number of output nodes in encoder is different. (2x z_dim for VAE, 1 for GAN)
 
-The following results can be reproduced with command:  
+The following results can be reproduced with command:
 ```
 python main.py --dataset mnist --gan_type <TYPE> --epoch 25 --batch_size 64
 ```
@@ -158,7 +159,7 @@ Results of CGAN is also given to compare images generated from CVAE and CGAN.
 
 #### Learned manifold
 
-The following results can be reproduced with command:  
+The following results can be reproduced with command:
 ```
 python main.py --dataset mnist --gan_type VAE --epoch 25 --batch_size 64 --dim_z 2
 ```
@@ -169,9 +170,9 @@ Please notice that dimension of noise-vector z is 2.
 VAE | <img src = 'assets/mnist_results/learned_manifold/VAE_epoch000_learned_manifold.png' height = '230px'> | <img src = 'assets/mnist_results/learned_manifold/VAE_epoch009_learned_manifold.png' height = '230px'> | <img src = 'assets/mnist_results/learned_manifold/VAE_epoch024_learned_manifold.png' height = '230px'>
 
 ### Results for fashion-mnist
-Comments on network architecture in mnist are also applied to here. 
+Comments on network architecture in mnist are also applied to here.
 
-The following results can be reproduced with command:  
+The following results can be reproduced with command:
 ```
 python main.py --dataset fashion-mnist --gan_type <TYPE> --epoch 40 --batch_size 64
 ```
@@ -198,7 +199,7 @@ Results of CGAN is also given to compare images generated from CVAE and CGAN.
 
 #### Learned manifold
 
-The following results can be reproduced with command:  
+The following results can be reproduced with command:
 ```
 python main.py --dataset fashion-mnist --gan_type VAE --epoch 25 --batch_size 64 --dim_z 2
 ```

data/get_mnist_data.py

@@ -0,0 +1,6 @@
+from tensorflow.contrib.learn.python.learn.datasets.mnist import read_data_sets
+import tensorflow as tf
+
+# run code in gan_env
+mnist = read_data_sets("mnist", one_hot=True)
+print(mnist)