jacobgil / keras-dcgan Goto Github PK

Keras implementation of Deep Convolutional Generative Adversarial Networks

Python 100.00%

deep-learning dcgan gan keras keras-dcgan artificial-images

keras-dcgan's Introduction

KERAS-DCGAN

Implementation of http://arxiv.org/abs/1511.06434 with the (awesome) keras library, for generating artificial images with deep learning.

This trains two adversarial deep learning models on real images, in order to produce artificial images that look real.

The generator model tries to produce images that look real and get a high score from the discriminator.

The discriminator model tries to tell apart between real images and artificial images from the generator.

This assumes theano ordering. You can still use this with tensorflow, by setting "image_dim_ordering": "th" in ~/.keras/keras.json (although this will be slower).

Usage

Training:

python dcgan.py --mode train --batch_size <batch_size>

python dcgan.py --mode train --path ~/images --batch_size 128

Image generation:

python dcgan.py --mode generate --batch_size <batch_size>

python dcgan.py --mode generate --batch_size <batch_size> --nice : top 5% images according to discriminator

python dcgan.py --mode generate --batch_size 128

Result

generated images :

train process :

keras-dcgan's People

Contributors

Stargazers

Watchers

Forkers

jfsantos amoliu wanjinchang countkisg whiteisclosing kastnerkyle ryotakatoh samueltenka arasharchor ffsouza drorhilman laoluzi ml-ai-nlp-ir hyzcn vyraun zhangqianhui benjamesbabala pursueorigin jmrinaldi wuqixiaobai el-cangrejo ghs2015 mvpcom locke-phan feherbalazs wassname kurnianggoro philkuz 309972460 adrianhust matthewwilfred minare k-du beckgom ahukui plok6325 gwaybio li099 bityangke dyelax mave5 crerecombinase zzhang1987 kryptonish sperfu hpjmlgy lunardog zhangyuancv recluze yunitata lukasmosser karimpedia nguizard snakeroot91 notbanker buptss praneetdutta asnelt kyuhwanjung zdenekolar wuatanabe mridulbirla gabrielnan stockedge asadi8 jskdr cjliux qian2729 arkwave shaform duolinwang gkgsd arashmh icemic iamgroot42 fanhuaandluomu superresolution simonsleo rparrapy gyom ankanbansal zhaoj9014 david-woo-instacart hardbyte hitfishking benjamwhite davidmcdonald1993 singhvarun404 pangyuteng ranchyliu wolfhu dodolimes qjin2016 daniel451 blackhc jiameng1010 ldoublev renato145 kentchun33333 wilsoncernwq

keras-dcgan's Issues

Why do you change d.trainable twice after d.compile during train?

It seems you can only change one model's trainable before model.compile, if you compile one model, your change of trainable will not work. So, why do you change model.trainable twice during train :
x= np.concatenate((image_batch, generated_images))
y = [1] * BATCH_SIZE + [0] * BATCH_SIZE
d_loss = d.train_on_batch(X, y)
print("batch %d d_loss : %f" % (index, d_loss))
noise = np.random.uniform(-1, 1, (BATCH_SIZE, 100))
d.trainable = False #'here'
g_loss = d_on_g.train_on_batch(noise, [1] * BATCH_SIZE)
d.trainable = True #'here'
print("batch %d g_loss : %f" % (index, g_loss))

Slow to converge

When I ran this, after 4 epochs G appeared to be farther and farther from converging (currently g_loss~=3) while D does appear to be converging slowly. Can you post a tensorboard screenshot of the convergence behavior?

TypeError: init() missing 1 required positional argument: 'units'

python dcgan.py --mode train --batch_size

I am loading the mnist dataset from a local pickle file, and this error is occuring. Debugging now, and will update as I learn more

//anaconda/envs/py35/lib/python3.5/importlib/_bootstrap.py:222: RuntimeWarning: compiletime version 3.6 of module 'tensorflow.python.framework.fast_tensor_util' does not match runtime version 3.5
return f(*args, **kwds)
//anaconda/envs/py35/lib/python3.5/importlib/_bootstrap.py:222: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88
return f(*args, **kwds)
//anaconda/envs/py35/lib/python3.5/importlib/_bootstrap.py:222: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88
return f(*args, **kwds)
//anaconda/envs/py35/lib/python3.5/importlib/_bootstrap.py:222: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88
return f(*args, **kwds)
Traceback (most recent call last):
File "DCGAN.py", line 167, in
train(BATCH_SIZE=args.batch_size)
File "DCGAN.py", line 91, in train
g = generator_model()
File "DCGAN.py", line 22, in generator_model
model.add(Dense(input_dim=100, output_dim=1024))

could it be my TF version? It shouldn't be...

ImportError: cannot import name 'BatchNormalization' from 'keras.layers.normalization'

Hi,
I kept getting the below error when trying to execute dcgan.py file, when trying to execute in Colab

ImportError: cannot import name 'BatchNormalization' from 'keras.layers.normalization'

To over come this replace in the initial inclusions from line no 1 to 8 as below with Tensorflow

For line no 1 from tensorflow.keras.models import Sequential
For line no 2:8 replace it with from tensorflow.keras.layers import ( BatchNormalization, SeparableConv2D, MaxPooling2D, Activation, Flatten, Dropout, Dense, UpSampling2D,Reshape,Conv2D )
For line no 9 replace with from tensorflow.keras.optimizers import SGD
For line no 10 replace with from tensorflow.keras.datasets import mnist

Hope this helps!

Regards,
Mritula

ValueError: Negative dimension size caused by subtracting 2 from 1

When I type python dcgan.py --mode train --batch_size 32, I will see this
File "dcgan.py", line 167, in
train(BATCH_SIZE=args.batch_size)
File "dcgan.py", line 80, in train
discriminator = discriminator_model()
File "dcgan.py", line 41, in discriminator_model
model.add(MaxPooling2D(pool_size=(2, 2)))
File "/usr/local/lib/python2.7/site-packages/keras/models.py", line 312, in add
output_tensor = layer(self.outputs[0])
File "/usr/local/lib/python2.7/site-packages/keras/engine/topology.py", line 514, in call
self.add_inbound_node(inbound_layers, node_indices, tensor_indices)
File "/usr/local/lib/python2.7/site-packages/keras/engine/topology.py", line 572, in add_inbound_node
Node.create_node(self, inbound_layers, node_indices, tensor_indices)
File "/usr/local/lib/python2.7/site-packages/keras/engine/topology.py", line 149, in create_node
output_tensors = to_list(outbound_layer.call(input_tensors[0], mask=input_masks[0]))
File "/usr/local/lib/python2.7/site-packages/keras/layers/pooling.py", line 162, in call
dim_ordering=self.dim_ordering)
File "/usr/local/lib/python2.7/site-packages/keras/layers/pooling.py", line 212, in _pooling_function
border_mode, dim_ordering, pool_mode='max')
File "/usr/local/lib/python2.7/site-packages/keras/backend/tensorflow_backend.py", line 1761, in pool2d
x = tf.nn.max_pool(x, pool_size, strides, padding=padding)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/nn_ops.py", line 850, in max_pool
name=name)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/ops/gen_nn_ops.py", line 1440, in _max_pool
data_format=data_format, name=name)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.py", line 749, in apply_op
op_def=op_def)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2382, in create_op
set_shapes_for_outputs(ret)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1783, in set_shapes_for_outputs
shapes = shape_func(op)
File "/usr/local/lib/python2.7/site-packages/tensorflow/python/framework/common_shapes.py", line 596, in call_cpp_shape_fn
raise ValueError(err.message)
ValueError: Negative dimension size caused by subtracting 2 from 1
I wonder how to fix it.

127.5

hi,what's the use of 127.5 in this code：
X_train = (X_train.astype(np.float32) - 127.5)/127.5

to scale the input datas? If yes ,than how do you calculate 127.5 , is it just match minist data?

Loss Function

In the paper GAN,the loss funtion has a form of minmaxV(G,D),while in your code the loss is 'binary_crossentropy',I'm little confused about this,could you please explain this a little bit?

Activation function of generator and discriminator?

Hello.

In dcgan.py generator_model(), the model uses tanh for not only last layer but every layer.
Original paper said, it's recommended to use ReLU for every layer except the last one.
Same difference exsist in discriminator_model as well.

Do you have some specific reason that you build these differently?

Thank you for sharing your code, though.

Train on custom datasets

Hi,
Is it possible to train on our own dataset? if yes then how ? I am using this command python dcgan.py --mode train --path ~/images --batch_size 128
after placing images in the /images folder but it throw the error as there is no argument --path , I also checked your python file it does not include --path argument.

Error when training discriminator on generator

So, that. I get the following error:
theano.gof.fg.MissingInputError: ("An input of the graph, used to compute DimShuffle{0,1,2,x}(convolution1d_input_1), was not provided and not given a value.Use the Theano flag exception_verbosity='high',for more information on this error.", convolution1d_input_1)

I dug around and found that when training the whole thing, the discriminator had .trainable=True. So, even though it was compiled with False, when you turn it on afterwards it becomes trainable, and then the Theano graph is not well defined.

I was able to fix this by setting discriminator.trainable=True and recompiling it before its training, and afterwards setting it to False and recompiling the discriminator_on_generator.

Now I need to understand why the discriminator converges very fast while the generator diverges. Any ideas on this? I am not using images, but midi files expressed as vector sequences.

Dataset for Training

I trained the model on various datasets having more then 20k images, but even after several epochs i'm not getting the desired results.

Can I know the dataset on which this model has been trained and tested?

Thank You

Potential label error?

since we're training discriminator_on_generator with noise, shouldnt the corresponding label data be a string of 0s instead of 1s?

for i in range(BATCH_SIZE):
   noise[i, :] = np.random.uniform(-1, 1, 100)
g_loss = discriminator_on_generator.train_on_batch(noise, [1] * BATCH_SIZE)  # shouldnt this be [0] * BATCH_SIZE?

thanks!

ValueError: all the input arrays must have same number of dimensions

When I'm trying to test this model with my own images (not MNIST) I get this error:

I load the images via scipy then reshape these to 28 by 28 pixels.

I use batch_size of 32, number of labels is 5

What's going on here, man? What am I doing wrong? 😞

What's the meaning of this code?

In the
def train(BATCH_SIZE):
discriminator_on_generator = \ generator_containing_discriminator(generator, discriminator)

what is discriminator_on_generator?Is it a model?

Python error 'int' object has no attribute 'shape'

I kept getting an error " Python error 'int' object has no attribute 'shape'" reflecting to the following line from def train in the dcgan.py file

g_loss = d_on_g.train_on_batch(noise, [1] * BATCH_SIZE))

Solution:
Changing the line to the below worked well:
g_loss = d_on_g.train_on_batch(noise,np.array([1] * BATCH_SIZE))

Correct me if this is wrong!

Hope this is of some help!

Thanks,
Mritula

ValueError: Negative dimension size caused by subtracting 2 from 1 for 'max_pooling2d_1/MaxPool'

When I try to run: python dcgan.py --mode train --batch_size 100 I get the following:


Using TensorFlow backend.
dcgan.py:41: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(64, (5, 5), input_shape=(1, 28, 28..., padding="same")`
  input_shape=(1, 28, 28)))
Traceback (most recent call last):
  File "/home/marija/.local/lib/python3.5/site-packages/tensorflow/python/framework/common_shapes.py", line 671, in _call_cpp_shape_fn_impl
    input_tensors_as_shapes, status)
  File "/home/marija/anaconda3/envs/tensorflow-gpu-3.5/lib/python3.5/contextlib.py", line 66, in __exit__
    next(self.gen)
  File "/home/marija/.local/lib/python3.5/site-packages/tensorflow/python/framework/errors_impl.py", line 466, in raise_exception_on_not_ok_status
    pywrap_tensorflow.TF_GetCode(status))
tensorflow.python.framework.errors_impl.InvalidArgumentError: Negative dimension size caused by subtracting 2 from 1 for 'max_pooling2d_1/MaxPool' (op: 'MaxPool') with input shapes: [?,1,28,64].

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "dcgan.py", line 169, in <module>
    train(BATCH_SIZE=args.batch_size)
  File "dcgan.py", line 82, in train
    discriminator = discriminator_model()
  File "dcgan.py", line 43, in discriminator_model
    model.add(MaxPooling2D(pool_size=(2, 2)))
  File "/home/marija/.local/lib/python3.5/site-packages/keras/models.py", line 466, in add
    output_tensor = layer(self.outputs[0])
  File "/home/marija/.local/lib/python3.5/site-packages/keras/engine/topology.py", line 585, in __call__
    output = self.call(inputs, **kwargs)
  File "/home/marija/.local/lib/python3.5/site-packages/keras/layers/pooling.py", line 154, in call
    data_format=self.data_format)
  File "/home/marija/.local/lib/python3.5/site-packages/keras/layers/pooling.py", line 217, in _pooling_function
    pool_mode='max')
  File "/home/marija/.local/lib/python3.5/site-packages/keras/backend/tensorflow_backend.py", line 3245, in pool2d
    x = tf.nn.max_pool(x, pool_size, strides, padding=padding)
  File "/home/marija/.local/lib/python3.5/site-packages/tensorflow/python/ops/nn_ops.py", line 1821, in max_pool
    name=name)
  File "/home/marija/.local/lib/python3.5/site-packages/tensorflow/python/ops/gen_nn_ops.py", line 1638, in _max_pool
    data_format=data_format, name=name)
  File "/home/marija/.local/lib/python3.5/site-packages/tensorflow/python/framework/op_def_library.py", line 768, in apply_op
    op_def=op_def)
  File "/home/marija/.local/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 2338, in create_op
    set_shapes_for_outputs(ret)
  File "/home/marija/.local/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 1719, in set_shapes_for_outputs
    shapes = shape_func(op)
  File "/home/marija/.local/lib/python3.5/site-packages/tensorflow/python/framework/ops.py", line 1669, in call_with_requiring
    return call_cpp_shape_fn(op, require_shape_fn=True)
  File "/home/marija/.local/lib/python3.5/site-packages/tensorflow/python/framework/common_shapes.py", line 610, in call_cpp_shape_fn
    debug_python_shape_fn, require_shape_fn)
  File "/home/marija/.local/lib/python3.5/site-packages/tensorflow/python/framework/common_shapes.py", line 676, in _call_cpp_shape_fn_impl
    raise ValueError(err.message)
ValueError: Negative dimension size caused by subtracting 2 from 1 for 'max_pooling2d_1/MaxPool' (op: 'MaxPool') with input shapes: [?,1,28,64].

Any ideas what is going wrong here?

ValueError: Variable batch_normalization_1/moving_mean/biased already exists

I am using tensorflow backend and updated the keras.json file as -

{
    "floatx": "float32",
    "epsilon": 1e-07,
    "backend": "tensorflow",
    "image_data_format": "channels_last"
    "image_dim_ordering": "th"
}

This is the error that i am getting -

Using TensorFlow backend.
test2.py:31: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(input_dim=100, units=1024)`
  model.add(Dense(input_dim=100, output_dim=1024))
Traceback (most recent call last):
  File "test2.py", line 177, in <module>
    train(BATCH_SIZE=args.batch_size)
  File "test2.py", line 102, in train
    d_on_g = generator_containing_discriminator(g, d)
  File "test2.py", line 71, in generator_containing_discriminator
    model.add(g)
  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\models.py", line 467, in add
    layer(x)
  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\engine\topology.py", line 619, in __call__
    output = self.call(inputs, **kwargs)
  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\models.py", line 549, in call
    return self.model.call(inputs, mask)
  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\engine\topology.py", line 2085, in call
    output_tensors, _, _ = self.run_internal_graph(inputs, masks)
  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\engine\topology.py", line 2236, in run_internal_graph
    output_tensors = _to_list(layer.call(computed_tensor, **kwargs))
  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\layers\normalization.py", line 193, in call
    self.momentum),
  File "C:\ProgramData\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py", line 1004, in moving_average_up
date
    x, value, momentum, zero_debias=True)
  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\training\moving_averages.py", line 70, in assign_mo
ving_average
    update_delta = _zero_debias(variable, value, decay)
  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\training\moving_averages.py", line 180, in _zero_de
bias
    "biased", initializer=biased_initializer, trainable=False)
  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 1065, in get_variable
    use_resource=use_resource, custom_getter=custom_getter)
  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 962, in get_variable
    use_resource=use_resource, custom_getter=custom_getter)
  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 367, in get_variable
    validate_shape=validate_shape, use_resource=use_resource)
  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 352, in _true_getter
    use_resource=use_resource)
  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 664, in _get_single_va
riable
    name, "".join(traceback.format_list(tb))))
ValueError: Variable batch_normalization_1/moving_mean/biased already exists, disallowed. Did you mean to set reuse=True
 in VarScope? Originally defined at:

  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\framework\ops.py", line 1204, in __init__
    self._traceback = self._graph._extract_stack()  # pylint: disable=protected-access
  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\framework\ops.py", line 2630, in create_op
    original_op=self._default_original_op, op_def=op_def)
  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\framework\op_def_library.py", line 767, in apply_op

    op_def=op_def)

Any idea what i might be doing wrong?

Other dataset

@jacobgil @jusjusjus @jacobsagivtech @laoluzi How can use this code for another dataset or own dataset and how access the discriminator network as feature extractor of image ?

Assertion Error

I get an Assertion Error when compiling the model, haven't changed anything in the code.

Do you know that the problem might be?

Traceback (most recent call last):
  File "dcgan.py", line 155, in <module>
    train(path = args.path, BATCH_SIZE = args.batch_size)
  File "dcgan.py", line 87, in train
    discriminator_on_generator.compile(loss='binary_crossentropy', optimizer=adam)
  File "/usr/local/lib/python3.4/dist-packages/keras/models.py", line 467, in compile
    self.y_train = self.get_output(train=True)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/containers.py", line 128, in get_output
    return self.layers[-1].get_output(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/containers.py", line 128, in get_output
    return self.layers[-1].get_output(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 679, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 970, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 842, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/advanced_activations.py", line 28, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/normalization.py", line 71, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/convolutional.py", line 312, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/advanced_activations.py", line 28, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/normalization.py", line 71, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/convolutional.py", line 312, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/advanced_activations.py", line 28, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/normalization.py", line 71, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/convolutional.py", line 312, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/normalization.py", line 71, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/advanced_activations.py", line 28, in get_output
    X = self.get_input(train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/core.py", line 175, in get_input
    previous_output = self.previous.get_output(train=train)
  File "/usr/local/lib/python3.4/dist-packages/keras/layers/convolutional.py", line 317, in get_output
    filter_shape=self.W_shape)
  File "/usr/local/lib/python3.4/dist-packages/keras/backend/theano_backend.py", line 596, in conv2d
    assert(strides == (1, 1))
AssertionError

noise sampling

is there a reason you sampled from an uniform distribution (as opposed to a normal distribution)? just wondering if this was intentional or not!

Why using different noise for training `d` and `d_on_g`

Hi
In line 96 we are defining noise which is used to generate images and train Discriminator d.
In line 108 we are again defining new noise which is used to train Generator d_on_g

Could you please explain why are we not using the noise defined in line 96? Is it because if we use the same starting point to train d and d_on_g it may overfit?

Thanks

where is the generator weights?

OSError: Unable to open file (unable to open file: name = 'generator'
so I cant't find the weight in your code line 121 : g.load_weights('generator')

Do you have a trained model? Thanks

doesn't work out of the box

Hello,

I tried to launch this project locally. I set my keras to use theano, python = 2.7; its ubuntu.

Traceback (most recent call last):
File "dcgun.py", line 167, in
train(BATCH_SIZE=args.batch_size)
File "dcgun.py", line 80, in train
discriminator = discriminator_model()
File "dcgun.py", line 46, in discriminator_model
model.add(Dense(1024))
File "/usr/local/lib/python2.7/dist-packages/keras/models.py", line 312, in add
output_tensor = layer(self.outputs[0])
File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 487, in call
self.build(input_shapes[0])
File "/usr/local/lib/python2.7/dist-packages/keras/layers/core.py", line 695, in build
name='{}_W'.format(self.name))
File "/usr/local/lib/python2.7/dist-packages/keras/initializations.py", line 59, in glorot_uniform
return uniform(shape, s, name=name)
File "/usr/local/lib/python2.7/dist-packages/keras/initializations.py", line 32, in uniform
return K.random_uniform_variable(shape, -scale, scale, name=name)
File "/usr/local/lib/python2.7/dist-packages/keras/backend/theano_backend.py", line 140, in random_uniform_variable
return variable(np.random.uniform(low=low, high=high, size=shape),
File "mtrand.pyx", line 1565, in mtrand.RandomState.uniform (numpy/random/mtrand/mtrand.c:17319)
OverflowError: Range exceeds valid bounds

Disrepancy in architecture in paper vs implemented version

To me it seems that the architecture in the referenced paper is quite different from the one that is implemented right now.

In particular:

the use of relu and leakyrely activations.
batchnormalisation between the layers
striding instead of pooling

BTW: I really like this initiative (looking to use adversarial nets within keras quite soon.)

AttributeError: 'int' object has no attribute 'ndim'

Keras2.1.3
Python 2.7
Thenao 1.0.1

root@localhost:~/zya/keras-dcgan# python dcgan.py --mode train
/usr/local/lib/python2.7/dist-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters
Using Theano backend.
dcgan.py:19: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(units=1024, input_dim=100)`
  model.add(Dense(input_dim=100, output_dim=1024))
('Epoch is', 0)
('Number of batches', 468)
Traceback (most recent call last):
  File "dcgan.py", line 160, in <module>
    train(BATCH_SIZE=args.batch_size)
  File "dcgan.py", line 106, in train
    d_loss = d.train_on_batch(X, y)
  File "/usr/local/lib/python2.7/dist-packages/keras/models.py", line 1069, in train_on_batch
    class_weight=class_weight)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1843, in train_on_batch
    check_batch_axis=True)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1430, in _standardize_user_data
    exception_prefix='target')
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 70, in _standardize_input_data
    data = [np.expand_dims(x, 1) if x is not None and x.ndim == 1 else x for x in data]
AttributeError: 'int' object has no attribute 'ndim'

The code does not inhibit the learning process of the discriminator

The line discriminator.trainable = False does not stop the discriminator from learning. Replace that types of line with a call at the following function:

def make_trainable(net, val):
     net.trainable = val
     for l in net.layers:
         l.trainable = val

No License.

I cannot see the license on this repository. Please choose appropriate license if its okay for others to derive and use your wonderful work[1].
Thanks.
[1]. https://help.github.com/articles/open-source-licensing/#what-happens-if-i-dont-choose-a-license

Why did not use Conv2DTranspose rather than Conv2D in the generator

the related paper all say they use Conv2DTranspose