The Wasserstein Generative Adversarial Network, or Wasserstein GAN, is an extension to the generative adversarial network that both improves the stability when training the model and provides a loss function that correlates with the quality of generated images.

The development of the WGAN has a dense mathematical motivation, although in practice requires only a few minor modifications to the established standard deep convolutional generative adversarial network, or DCGAN.

In this tutorial, you will discover how to implement the Wasserstein generative adversarial network from scratch.

After completing this tutorial, you will know:

- The differences between the standard deep convolutional GAN and the new Wasserstein GAN.
- How to implement the specific details of the Wasserstein GAN from scratch.
- How to develop a WGAN for image generation and interpret the dynamic behavior of the model.

Discover how to develop DCGANs, conditional GANs, Pix2Pix, CycleGANs, and more with Keras in my new GANs book, with 29 step-by-step tutorials and full source code.

Let’s get started.

## Tutorial Overview

This tutorial is divided into three parts; they are:

- Wasserstein Generative Adversarial Network
- Wasserstein GAN Implementation Details
- How to Train a Wasserstein GAN Model

## Wasserstein Generative Adversarial Network

The Wasserstein GAN, or WGAN for short, was introduced by Martin Arjovsky, et al. in their 2017 paper titled “Wasserstein GAN.”

It is an extension of the GAN that seeks an alternate way of training the generator model to better approximate the distribution of data observed in a given training dataset.

Instead of using a discriminator to classify or predict the probability of generated images as being real or fake, the WGAN changes or replaces the discriminator model with a critic that scores the realness or fakeness of a given image.

This change is motivated by a theoretical argument that training the generator should seek a minimization of the distance between the distribution of the data observed in the training dataset and the distribution observed in generated examples.

The benefit of the WGAN is that the training process is more stable and less sensitive to model architecture and choice of hyperparameter configurations. Perhaps most importantly, the loss of the discriminator appears to relate to the quality of images created by the generator.

## Wasserstein GAN Implementation Details

Although the theoretical grounding for the WGAN is dense, the implementation of a WGAN requires a few minor changes to the standard Deep Convolutional GAN, or DCGAN.

The image below provides a summary of the main training loop for training a WGAN, taken from the paper. Note the listing of recommended hyperparameters used in the model.

The differences in implementation for the WGAN are as follows:

- Use a linear activation function in the output layer of the critic model (instead of sigmoid).
- Use -1 labels for real images and 1 labels for fake images (instead of 1 and 0).
- Use Wasserstein loss to train the critic and generator models.
- Constrain critic model weights to a limited range after each mini batch update (e.g. [-0.01,0.01]).
- Update the critic model more times than the generator each iteration (e.g. 5).
- Use the RMSProp version of gradient descent with a small learning rate and no momentum (e.g. 0.00005).

Using the standard DCGAN model as a starting point, let’s take a look at each of these implementation details in turn.

### Want to Develop GANs from Scratch?

Take my free 7-day email crash course now (with sample code).

Click to sign-up and also get a free PDF Ebook version of the course.

### 1. Linear Activation in Critic Output Layer

The DCGAN uses the sigmoid activation function in the output layer of the discriminator to predict the likelihood of a given image being real.

In the WGAN, the critic model requires a linear activation to predict the score of “*realness*” for a given image.

This can be achieved by setting the ‘*activation*‘ argument to ‘*linear*‘ in the output layer of the critic model.

1 2 3 |
# define output layer of the critic model ... model.add(Dense(1, activation='linear')) |

The linear activation is the default activation for a layer, so we can, in fact, leave the activation unspecified to achieve the same result.

1 2 3 |
# define output layer of the critic model ... model.add(Dense(1)) |

### 2. Class Labels for Real and Fake Images

The DCGAN uses the class 0 for fake images and class 1 for real images, and these class labels are used to train the GAN.

In the DCGAN, these are precise labels that the discriminator is expected to achieve. The WGAN does not have precise labels for the critic. Instead, it encourages the critic to output scores that are different for real and fake images.

This is achieved via the Wasserstein function that cleverly makes use of positive and negative class labels.

The WGAN can be implemented where -1 class labels are used for real images and +1 class labels are used for fake or generated images.

This can be achieved using the ones() NumPy function.

For example:

1 2 3 4 5 6 |
... # generate class labels, -1 for 'real' y = -ones((n_samples, 1)) ... # create class labels with 1.0 for 'fake' y = ones((n_samples, 1)) |

### 3. Wasserstein Loss Function

The DCGAN trains the discriminator as a binary classification model to predict the probability that a given image is real.

To train this model, the discriminator is optimized using the binary cross entropy loss function. The same loss function is used to update the generator model.

The primary contribution of the WGAN model is the use of a new loss function that encourages the discriminator to predict a score of how real or fake a given input looks. This transforms the role of the discriminator from a classifier into a critic for scoring the realness or fakeness of images, where the difference between the scores is as large as possible.

We can implement the Wasserstein loss as a custom function in Keras that calculates the average score for real or fake images.

The score is maximizing for real examples and minimizing for fake examples. Given that stochastic gradient descent is a minimization algorithm, we can multiply the class label by the mean score (e.g. -1 for real and 1 for fake which as no effect), which ensures that the loss for real and fake images is minimizing to the network.

An efficient implementation of this loss function for Keras is listed below.

1 2 3 4 5 |
from keras import backend # implementation of wasserstein loss def wasserstein_loss(y_true, y_pred): return backend.mean(y_true * y_pred) |

This loss function can be used to train a Keras model by specifying the function name when compiling the model.

For example:

1 2 3 |
... # compile the model model.compile(loss=wasserstein_loss, ...) |

### 4. Critic Weight Clipping

The DCGAN does not use any gradient clipping, although the WGAN requires gradient clipping for the critic model.

We can implement weight clipping as a Keras constraint.

This is a class that must extend the *Constraint* class and define an implementation of the *__call__()* function for applying the operation and the *get_config()* function for returning any configuration.

We can also define an *__init__()* function to set the configuration, in this case, the symmetrical size of the bounding box for the weight hypercube, e.g. 0.01.

The *ClipConstraint* class is defined below.

1 2 3 4 5 6 7 8 9 10 11 12 13 |
# clip model weights to a given hypercube class ClipConstraint(Constraint): # set clip value when initialized def __init__(self, clip_value): self.clip_value = clip_value # clip model weights to hypercube def __call__(self, weights): return backend.clip(weights, -self.clip_value, self.clip_value) # get the config def get_config(self): return {'clip_value': self.clip_value} |

To use the constraint, the class can be constructed, then used in a layer by setting the *kernel_constraint* argument; for example:

1 2 3 4 5 6 |
... # define the constraint const = ClipConstraint(0.01) ... # use the constraint in a layer model.add(Conv2D(..., kernel_constraint=const)) |

The constraint is only required when updating the critic model.

### 5. Update Critic More Than Generator

In the DCGAN, the generator and the discriminator model must be updated in equal amounts.

Specifically, the discriminator is updated with a half batch of real and a half batch of fake samples each iteration, whereas the generator is updated with a single batch of generated samples.

For example:

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 |
... # main gan training loop for i in range(n_steps): # update the discriminator # get randomly selected 'real' samples X_real, y_real = generate_real_samples(dataset, half_batch) # update critic model weights c_loss1 = c_model.train_on_batch(X_real, y_real) # generate 'fake' examples X_fake, y_fake = generate_fake_samples(g_model, latent_dim, half_batch) # update critic model weights c_loss2 = c_model.train_on_batch(X_fake, y_fake) # update generator # prepare points in latent space as input for the generator X_gan = generate_latent_points(latent_dim, n_batch) # create inverted labels for the fake samples y_gan = ones((n_batch, 1)) # update the generator via the critic's error g_loss = gan_model.train_on_batch(X_gan, y_gan) |

In the WGAN model, the critic model must be updated more than the generator model.

Specifically, a new hyperparameter is defined to control the number of times that the critic is updated for each update to the generator model, called n_critic, and is set to 5.

This can be implemented as a new loop within the main GAN update loop; for example:

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 |
... # main gan training loop for i in range(n_steps): # update the critic for _ in range(n_critic): # get randomly selected 'real' samples X_real, y_real = generate_real_samples(dataset, half_batch) # update critic model weights c_loss1 = c_model.train_on_batch(X_real, y_real) # generate 'fake' examples X_fake, y_fake = generate_fake_samples(g_model, latent_dim, half_batch) # update critic model weights c_loss2 = c_model.train_on_batch(X_fake, y_fake) # update generator # prepare points in latent space as input for the generator X_gan = generate_latent_points(latent_dim, n_batch) # create inverted labels for the fake samples y_gan = ones((n_batch, 1)) # update the generator via the critic's error g_loss = gan_model.train_on_batch(X_gan, y_gan) |

### 6. Use RMSProp Stochastic Gradient Descent

The DCGAN uses the Adam version of stochastic gradient descent with a small learning rate and modest momentum.

The WGAN recommends the use of RMSProp instead, with a small learning rate of 0.00005.

This can be implemented in Keras when the model is compiled. For example:

1 2 3 4 |
... # compile model opt = RMSprop(lr=0.00005) model.compile(loss=wasserstein_loss, optimizer=opt) |

## How to Train a Wasserstein GAN Model

Now that we know the specific implementation details for the WGAN, we can implement the model for image generation.

In this section, we will develop a WGAN to generate a single handwritten digit (‘7’) from the MNIST dataset. This is a good test problem for the WGAN as it is a small dataset requiring a modest mode that is quick to train.

The first step is to define the models.

The critic model takes as input one 28×28 grayscale image and outputs a score for the realness or fakeness of the image. It is implemented as a modest convolutional neural network using best practices for DCGAN design such as using the LeakyReLU activation function with a slope of 0.2, batch normalization, and using a 2×2 stride to downsample.

The critic model makes use of the new ClipConstraint weight constraint to clip model weights after mini-batch updates and is optimized using the custom *wasserstein_loss()* function, the RMSProp version of stochastic gradient descent with a learning rate of 0.00005.

The *define_critic()* function below implements this, defining and compiling the critic model and returning it. The input shape of the image is parameterized as a default function argument to make it clear.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 |
# define the standalone critic model def define_critic(in_shape=(28,28,1)): # weight initialization init = RandomNormal(stddev=0.02) # weight constraint const = ClipConstraint(0.01) # define model model = Sequential() # downsample to 14x14 model.add(Conv2D(64, (4,4), strides=(2,2), padding='same', kernel_initializer=init, kernel_constraint=const, input_shape=in_shape)) model.add(BatchNormalization()) model.add(LeakyReLU(alpha=0.2)) # downsample to 7x7 model.add(Conv2D(64, (4,4), strides=(2,2), padding='same', kernel_initializer=init, kernel_constraint=const)) model.add(BatchNormalization()) model.add(LeakyReLU(alpha=0.2)) # scoring, linear activation model.add(Flatten()) model.add(Dense(1)) # compile model opt = RMSprop(lr=0.00005) model.compile(loss=wasserstein_loss, optimizer=opt) return model |

The generator model takes as input a point in the latent space and outputs a single 28×28 grayscale image.

This is achieved by using a fully connected layer to interpret the point in the latent space and provide sufficient activations that can be reshaped into many copies (in this case, 128) of a low-resolution version of the output image (e.g. 7×7). This is then upsampled two times, doubling the size and quadrupling the area of the activations each time using transpose convolutional layers.

The model uses best practices such as the LeakyReLU activation, a kernel size that is a factor of the stride size, and a hyperbolic tangent (tanh) activation function in the output layer.

The *define_generator()* function below defines the generator model but intentionally does not compile it as it is not trained directly, then returns the model. The size of the latent space is parameterized as a function argument.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 |
# define the standalone generator model def define_generator(latent_dim): # weight initialization init = RandomNormal(stddev=0.02) # define model model = Sequential() # foundation for 7x7 image n_nodes = 128 * 7 * 7 model.add(Dense(n_nodes, kernel_initializer=init, input_dim=latent_dim)) model.add(LeakyReLU(alpha=0.2)) model.add(Reshape((7, 7, 128))) # upsample to 14x14 model.add(Conv2DTranspose(128, (4,4), strides=(2,2), padding='same', kernel_initializer=init)) model.add(BatchNormalization()) model.add(LeakyReLU(alpha=0.2)) # upsample to 28x28 model.add(Conv2DTranspose(128, (4,4), strides=(2,2), padding='same', kernel_initializer=init)) model.add(BatchNormalization()) model.add(LeakyReLU(alpha=0.2)) # output 28x28x1 model.add(Conv2D(1, (7,7), activation='tanh', padding='same', kernel_initializer=init)) return model |

Next, a GAN model can be defined that combines both the generator model and the critic model into one larger model.

This larger model will be used to train the model weights in the generator, using the output and error calculated by the critic model. The critic model is trained separately, and as such, the model weights are marked as not trainable in this larger GAN model to ensure that only the weights of the generator model are updated. This change to the trainability of the critic weights only has an effect when training the combined GAN model, not when training the critic standalone.

This larger GAN model takes as input a point in the latent space, uses the generator model to generate an image, which is fed as input to the critic model, then output scored as real or fake. The model is fit using RMSProp with the custom *wasserstein_loss()* function.

The *define_gan()* function below implements this, taking the already defined generator and critic models as input.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 |
# define the combined generator and critic model, for updating the generator def define_gan(generator, critic): # make weights in the critic not trainable critic.trainable = False # connect them model = Sequential() # add generator model.add(generator) # add the critic model.add(critic) # compile model opt = RMSprop(lr=0.00005) model.compile(loss=wasserstein_loss, optimizer=opt) return model |

Now that we have defined the GAN model, we need to train it. But, before we can train the model, we require input data.

The first step is to load and scale the MNIST dataset. The whole dataset is loaded via a call to the *load_data()* Keras function, then a subset of the images is selected (about 5,000) that belongs to class 7, e.g. are a handwritten depiction of the number seven. Then the pixel values must be scaled to the range [-1,1] to match the output of the generator model.

The *load_real_samples()* function below implements this, returning the loaded and scaled subset of the MNIST training dataset ready for modeling.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 |
# load images def load_real_samples(): # load dataset (trainX, trainy), (_, _) = load_data() # select all of the examples for a given class selected_ix = trainy == 7 X = trainX[selected_ix] # expand to 3d, e.g. add channels X = expand_dims(X, axis=-1) # convert from ints to floats X = X.astype('float32') # scale from [0,255] to [-1,1] X = (X - 127.5) / 127.5 return X |

We will require one batch (or a half) batch of real images from the dataset each update to the GAN model. A simple way to achieve this is to select a random sample of images from the dataset each time.

The *generate_real_samples()* function below implements this, taking the prepared dataset as an argument, selecting and returning a random sample of images and their corresponding label for the critic, specifically target=-1 indicating that they are real images.

1 2 3 4 5 6 7 8 9 |
# select real samples def generate_real_samples(dataset, n_samples): # choose random instances ix = randint(0, dataset.shape[0], n_samples) # select images X = dataset[ix] # generate class labels, -1 for 'real' y = -ones((n_samples, 1)) return X, y |

Next, we need inputs for the generator model. These are random points from the latent space, specifically Gaussian distributed random variables.

The *generate_latent_points()* function implements this, taking the size of the latent space as an argument and the number of points required, and returning them as a batch of input samples for the generator model.

1 2 3 4 5 6 7 |
# generate points in latent space as input for the generator def generate_latent_points(latent_dim, n_samples): # generate points in the latent space x_input = randn(latent_dim * n_samples) # reshape into a batch of inputs for the network x_input = x_input.reshape(n_samples, latent_dim) return x_input |

Next, we need to use the points in the latent space as input to the generator in order to generate new images.

The *generate_fake_samples()* function below implements this, taking the generator model and size of the latent space as arguments, then generating points in the latent space and using them as input to the generator model.

The function returns the generated images and their corresponding label for the critic model, specifically target=1 to indicate they are fake or generated.

1 2 3 4 5 6 7 8 9 |
# use the generator to generate n fake examples, with class labels def generate_fake_samples(generator, latent_dim, n_samples): # generate points in latent space x_input = generate_latent_points(latent_dim, n_samples) # predict outputs X = generator.predict(x_input) # create class labels with 1.0 for 'fake' y = ones((n_samples, 1)) return X, y |

We need to record the performance of the model. Perhaps the most reliable way to evaluate the performance of a GAN is to use the generator to generate images, and then review and subjectively evaluate them.

The *summarize_performance()* function below takes the generator model at a given point during training and uses it to generate 100 images in a 10×10 grid, that are then plotted and saved to file. The model is also saved to file at this time, in case we would like to use it later to generate more images.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 |
# generate samples and save as a plot and save the model def summarize_performance(step, g_model, latent_dim, n_samples=100): # prepare fake examples X, _ = generate_fake_samples(g_model, latent_dim, n_samples) # scale from [-1,1] to [0,1] X = (X + 1) / 2.0 # plot images for i in range(10 * 10): # define subplot pyplot.subplot(10, 10, 1 + i) # turn off axis pyplot.axis('off') # plot raw pixel data pyplot.imshow(X[i, :, :, 0], cmap='gray_r') # save plot to file filename1 = 'generated_plot_%04d.png' % (step+1) pyplot.savefig(filename1) pyplot.close() # save the generator model filename2 = 'model_%04d.h5' % (step+1) g_model.save(filename2) print('>Saved: %s and %s' % (filename1, filename2)) |

In addition to image quality, it is a good idea to keep track of the loss and accuracy of the model over time.

The loss for the critic for real and fake samples can be tracked for each model update, as can the loss for the generator for each update. These can then be used to create line plots of loss at the end of the training run. The *plot_history()* function below implements this and saves the results to file.

1 2 3 4 5 6 7 8 9 |
# create a line plot of loss for the gan and save to file def plot_history(d1_hist, d2_hist, g_hist): # plot history pyplot.plot(d1_hist, label='crit_real') pyplot.plot(d2_hist, label='crit_fake') pyplot.plot(g_hist, label='gen') pyplot.legend() pyplot.savefig('plot_line_plot_loss.png') pyplot.close() |

We are now ready to fit the GAN model.

The model is fit for 10 training epochs, which is arbitrary, as the model begins generating plausible number-7 digits after perhaps the first few epochs. A batch size of 64 samples is used, and each training epoch involves 6,265/64, or about 97, batches of real and fake samples and updates to the model. The model is therefore trained for 10 epochs of 97 batches, or 970 iterations.

First, the critic model is updated for a half batch of real samples, then a half batch of fake samples, together forming one batch of weight updates. This is then repeated *n_critic* (5) times as required by the WGAN algorithm.

The generator is then updated via the composite GAN model. Importantly, the target label is set to -1 or real for the generated samples. This has the effect of updating the generator toward getting better at generating real samples on the next batch.

The *train()* function below implements this, taking the defined models, dataset, and size of the latent dimension as arguments and parameterizing the number of epochs and batch size with default arguments. The generator model is saved at the end of training.

The performance of the critic and generator models is reported each iteration. Sample images are generated and saved every epoch, and line plots of model performance are created and saved at the end of the run.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 |
# train the generator and critic def train(g_model, c_model, gan_model, dataset, latent_dim, n_epochs=10, n_batch=64, n_critic=5): # calculate the number of batches per training epoch bat_per_epo = int(dataset.shape[0] / n_batch) # calculate the number of training iterations n_steps = bat_per_epo * n_epochs # calculate the size of half a batch of samples half_batch = int(n_batch / 2) # lists for keeping track of loss c1_hist, c2_hist, g_hist = list(), list(), list() # manually enumerate epochs for i in range(n_steps): # update the critic more than the generator c1_tmp, c2_tmp = list(), list() for _ in range(n_critic): # get randomly selected 'real' samples X_real, y_real = generate_real_samples(dataset, half_batch) # update critic model weights c_loss1 = c_model.train_on_batch(X_real, y_real) c1_tmp.append(c_loss1) # generate 'fake' examples X_fake, y_fake = generate_fake_samples(g_model, latent_dim, half_batch) # update critic model weights c_loss2 = c_model.train_on_batch(X_fake, y_fake) c2_tmp.append(c_loss2) # store critic loss c1_hist.append(mean(c1_tmp)) c2_hist.append(mean(c2_tmp)) # prepare points in latent space as input for the generator X_gan = generate_latent_points(latent_dim, n_batch) # create inverted labels for the fake samples y_gan = -ones((n_batch, 1)) # update the generator via the critic's error g_loss = gan_model.train_on_batch(X_gan, y_gan) g_hist.append(g_loss) # summarize loss on this batch print('>%d, c1=%.3f, c2=%.3f g=%.3f' % (i+1, c1_hist[-1], c2_hist[-1], g_loss)) # evaluate the model performance every 'epoch' if (i+1) % bat_per_epo == 0: summarize_performance(i, g_model, latent_dim) # line plots of loss plot_history(c1_hist, c2_hist, g_hist) |

Now that all of the functions have been defined, we can create the models, load the dataset, and begin the training process.

1 2 3 4 5 6 7 8 9 10 11 12 13 |
# size of the latent space latent_dim = 50 # create the critic critic = define_critic() # create the generator generator = define_generator(latent_dim) # create the gan gan_model = define_gan(generator, critic) # load image data dataset = load_real_samples() print(dataset.shape) # train model train(generator, critic, gan_model, dataset, latent_dim) |

Tying all of this together, the complete example is listed below.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 |
# example of a wgan for generating handwritten digits from numpy import expand_dims from numpy import mean from numpy import ones from numpy.random import randn from numpy.random import randint from keras.datasets.mnist import load_data from keras import backend from keras.optimizers import RMSprop from keras.models import Sequential from keras.layers import Dense from keras.layers import Reshape from keras.layers import Flatten from keras.layers import Conv2D from keras.layers import Conv2DTranspose from keras.layers import LeakyReLU from keras.layers import BatchNormalization from keras.initializers import RandomNormal from keras.constraints import Constraint from matplotlib import pyplot # clip model weights to a given hypercube class ClipConstraint(Constraint): # set clip value when initialized def __init__(self, clip_value): self.clip_value = clip_value # clip model weights to hypercube def __call__(self, weights): return backend.clip(weights, -self.clip_value, self.clip_value) # get the config def get_config(self): return {'clip_value': self.clip_value} # calculate wasserstein loss def wasserstein_loss(y_true, y_pred): return backend.mean(y_true * y_pred) # define the standalone critic model def define_critic(in_shape=(28,28,1)): # weight initialization init = RandomNormal(stddev=0.02) # weight constraint const = ClipConstraint(0.01) # define model model = Sequential() # downsample to 14x14 model.add(Conv2D(64, (4,4), strides=(2,2), padding='same', kernel_initializer=init, kernel_constraint=const, input_shape=in_shape)) model.add(BatchNormalization()) model.add(LeakyReLU(alpha=0.2)) # downsample to 7x7 model.add(Conv2D(64, (4,4), strides=(2,2), padding='same', kernel_initializer=init, kernel_constraint=const)) model.add(BatchNormalization()) model.add(LeakyReLU(alpha=0.2)) # scoring, linear activation model.add(Flatten()) model.add(Dense(1)) # compile model opt = RMSprop(lr=0.00005) model.compile(loss=wasserstein_loss, optimizer=opt) return model # define the standalone generator model def define_generator(latent_dim): # weight initialization init = RandomNormal(stddev=0.02) # define model model = Sequential() # foundation for 7x7 image n_nodes = 128 * 7 * 7 model.add(Dense(n_nodes, kernel_initializer=init, input_dim=latent_dim)) model.add(LeakyReLU(alpha=0.2)) model.add(Reshape((7, 7, 128))) # upsample to 14x14 model.add(Conv2DTranspose(128, (4,4), strides=(2,2), padding='same', kernel_initializer=init)) model.add(BatchNormalization()) model.add(LeakyReLU(alpha=0.2)) # upsample to 28x28 model.add(Conv2DTranspose(128, (4,4), strides=(2,2), padding='same', kernel_initializer=init)) model.add(BatchNormalization()) model.add(LeakyReLU(alpha=0.2)) # output 28x28x1 model.add(Conv2D(1, (7,7), activation='tanh', padding='same', kernel_initializer=init)) return model # define the combined generator and critic model, for updating the generator def define_gan(generator, critic): # make weights in the critic not trainable critic.trainable = False # connect them model = Sequential() # add generator model.add(generator) # add the critic model.add(critic) # compile model opt = RMSprop(lr=0.00005) model.compile(loss=wasserstein_loss, optimizer=opt) return model # load images def load_real_samples(): # load dataset (trainX, trainy), (_, _) = load_data() # select all of the examples for a given class selected_ix = trainy == 7 X = trainX[selected_ix] # expand to 3d, e.g. add channels X = expand_dims(X, axis=-1) # convert from ints to floats X = X.astype('float32') # scale from [0,255] to [-1,1] X = (X - 127.5) / 127.5 return X # select real samples def generate_real_samples(dataset, n_samples): # choose random instances ix = randint(0, dataset.shape[0], n_samples) # select images X = dataset[ix] # generate class labels, -1 for 'real' y = -ones((n_samples, 1)) return X, y # generate points in latent space as input for the generator def generate_latent_points(latent_dim, n_samples): # generate points in the latent space x_input = randn(latent_dim * n_samples) # reshape into a batch of inputs for the network x_input = x_input.reshape(n_samples, latent_dim) return x_input # use the generator to generate n fake examples, with class labels def generate_fake_samples(generator, latent_dim, n_samples): # generate points in latent space x_input = generate_latent_points(latent_dim, n_samples) # predict outputs X = generator.predict(x_input) # create class labels with 1.0 for 'fake' y = ones((n_samples, 1)) return X, y # generate samples and save as a plot and save the model def summarize_performance(step, g_model, latent_dim, n_samples=100): # prepare fake examples X, _ = generate_fake_samples(g_model, latent_dim, n_samples) # scale from [-1,1] to [0,1] X = (X + 1) / 2.0 # plot images for i in range(10 * 10): # define subplot pyplot.subplot(10, 10, 1 + i) # turn off axis pyplot.axis('off') # plot raw pixel data pyplot.imshow(X[i, :, :, 0], cmap='gray_r') # save plot to file filename1 = 'generated_plot_%04d.png' % (step+1) pyplot.savefig(filename1) pyplot.close() # save the generator model filename2 = 'model_%04d.h5' % (step+1) g_model.save(filename2) print('>Saved: %s and %s' % (filename1, filename2)) # create a line plot of loss for the gan and save to file def plot_history(d1_hist, d2_hist, g_hist): # plot history pyplot.plot(d1_hist, label='crit_real') pyplot.plot(d2_hist, label='crit_fake') pyplot.plot(g_hist, label='gen') pyplot.legend() pyplot.savefig('plot_line_plot_loss.png') pyplot.close() # train the generator and critic def train(g_model, c_model, gan_model, dataset, latent_dim, n_epochs=10, n_batch=64, n_critic=5): # calculate the number of batches per training epoch bat_per_epo = int(dataset.shape[0] / n_batch) # calculate the number of training iterations n_steps = bat_per_epo * n_epochs # calculate the size of half a batch of samples half_batch = int(n_batch / 2) # lists for keeping track of loss c1_hist, c2_hist, g_hist = list(), list(), list() # manually enumerate epochs for i in range(n_steps): # update the critic more than the generator c1_tmp, c2_tmp = list(), list() for _ in range(n_critic): # get randomly selected 'real' samples X_real, y_real = generate_real_samples(dataset, half_batch) # update critic model weights c_loss1 = c_model.train_on_batch(X_real, y_real) c1_tmp.append(c_loss1) # generate 'fake' examples X_fake, y_fake = generate_fake_samples(g_model, latent_dim, half_batch) # update critic model weights c_loss2 = c_model.train_on_batch(X_fake, y_fake) c2_tmp.append(c_loss2) # store critic loss c1_hist.append(mean(c1_tmp)) c2_hist.append(mean(c2_tmp)) # prepare points in latent space as input for the generator X_gan = generate_latent_points(latent_dim, n_batch) # create inverted labels for the fake samples y_gan = -ones((n_batch, 1)) # update the generator via the critic's error g_loss = gan_model.train_on_batch(X_gan, y_gan) g_hist.append(g_loss) # summarize loss on this batch print('>%d, c1=%.3f, c2=%.3f g=%.3f' % (i+1, c1_hist[-1], c2_hist[-1], g_loss)) # evaluate the model performance every 'epoch' if (i+1) % bat_per_epo == 0: summarize_performance(i, g_model, latent_dim) # line plots of loss plot_history(c1_hist, c2_hist, g_hist) # size of the latent space latent_dim = 50 # create the critic critic = define_critic() # create the generator generator = define_generator(latent_dim) # create the gan gan_model = define_gan(generator, critic) # load image data dataset = load_real_samples() print(dataset.shape) # train model train(generator, critic, gan_model, dataset, latent_dim) |

Running the example is quick, taking approximately 10 minutes on modern hardware without a GPU.

Your specific results will vary given the stochastic nature of the learning algorithm. Nevertheless, the general structure of training should be very similar.

First, the loss of the critic and generator models is reported to the console each iteration of the training loop. Specifically, c1 is the loss of the critic on real examples, c2 is the loss of the critic in generated samples, and g is the loss of the generator trained via the critic.

The c1 scores are inverted as part of the loss function; this means if they are reported as negative, then they are really positive, and if they are reported as positive, they are really negative. The sign of the c2 scores is unchanged.

Recall that the Wasserstein loss seeks scores for real and fake that are more different during training. We can see this towards the end of the run, such as the final epoch where the *c1* loss for real examples is 5.338 (really -5.338) and the *c2* loss for fake examples is -14.260, and this separation of about 10 units is consistent at least for the prior few iterations.

We can also see that in this case, the model is scoring the loss of the generator at around 20. Again, recall that we update the generator via the critic model and treat the generated examples as real with the target of -1, therefore the score can be interpreted as a value around -20, close to the loss for fake samples.

1 2 3 4 5 6 7 8 9 10 11 |
... >961, c1=5.110, c2=-15.388 g=19.579 >962, c1=6.116, c2=-15.222 g=20.054 >963, c1=4.982, c2=-15.192 g=21.048 >964, c1=4.238, c2=-14.689 g=23.911 >965, c1=5.585, c2=-14.126 g=19.578 >966, c1=4.807, c2=-14.755 g=20.034 >967, c1=6.307, c2=-16.538 g=19.572 >968, c1=4.298, c2=-14.178 g=17.956 >969, c1=4.283, c2=-13.398 g=17.326 >970, c1=5.338, c2=-14.260 g=19.927 |

Line plots for loss are created and saved at the end of the run.

The plot shows the loss for the critic on real samples (blue), the loss for the critic on fake samples (orange), and the loss for the critic when updating the generator with fake samples (green).

There is one important factor when reviewing learning curves for the WGAN and that is the trend.

The benefit of the WGAN is that the loss correlates with generated image quality. Lower loss means better quality images, for a stable training process.

In this case, lower loss specifically refers to lower Wasserstein loss for generated images as reported by the critic (orange line). This sign of this loss is not inverted by the target label (e.g. the target label is +1.0), therefore, a well-performing WGAN should show this line trending down as the image quality of the generated model is increased.

In this case, more training seems to result in better quality generated images, with a major hurdle occurring around epoch 200-300 after which quality remains pretty good for the model.

Before and around this hurdle, image quality is poor; for example:

After this epoch, the WGAN continues to generate plausible handwritten digits.

## Further Reading

This section provides more resources on the topic if you are looking to go deeper.

### Papers

- Wasserstein GAN, 2017.
- Improved Training of Wasserstein GANs, 2017.

### API

- Keras Datasets API.
- Keras Sequential Model API
- Keras Convolutional Layers API
- How can I “freeze” Keras layers?
- MatplotLib API
- NumPy Random sampling (numpy.random) API
- NumPy Array manipulation routines

### Articles

- WassersteinGAN, GitHub.
- Wasserstein Generative Adversarial Networks (WGANS) Project, GitHub.
- Keras-GAN: Keras implementations of Generative Adversarial Networks, GitHub.
- Improved WGAN, keras-contrib Project, GitHub.

## Summary

In this tutorial, you discovered how to implement the Wasserstein generative adversarial network from scratch.

Specifically, you learned:

- The differences between the standard deep convolutional GAN and the new Wasserstein GAN.
- How to implement the specific details of the Wasserstein GAN from scratch.
- How to develop a WGAN for image generation and interpret the dynamic behavior of the model.

Do you have any questions?

Ask your questions in the comments below and I will do my best to answer.

Thank for your great tutorial

I am curious about 2 things:

1. What is exactly clipping weight does? (I read the source code of Keras for clip_by_value() function but I can’t figure out what exactly it does)

2. According to your previous article, I knew that Critic’s outcome is just a real number for optimizing the generator, as long as critic(fake) > critic(real). The critic is Wasserstein distance. Therefore, can Wasserstein loss use in common CNN replacing CNN loss function like MSE, PSNR? say, I am training a deblurring CNN using Wasserstein loss, if I minimize this loss value, then my CNN will converge? Am I correct?

Thank for reading my question

Good questions.

The weight clipping ensure weight values (model parameters) stay within a pre-defined bounds.

You can use MSE instead, but it is different. In fact, this is called a least-squares GAN and does not convergence, but does very well.

Thank for your response!

In my question 1, I still confuse about “stay within pre-defined bounds”. In my question 2, There is a misunderstanding.

Question 1: Assume I have weights like this W = [0.3 -2.5, 4.5, 0.05, -0.02] then what is the result of clip(W, (-0.1, 0.1)) ?

Question 2: I know least-squares GAN, my question is: “how to adopt Wasserstein loss metric as loss function in NORMAL architecture CNN (not GAN architecture)?”. I did search on the internet to find the answer but there is no study adopting this Wasserstein Distance as their loss function in normal CNN. I didn’t figure out the reason why they didn’t use it.

I am so sorry for my bad English

Hope you get what I mean.

Thank you

Good questions.

The clipping is defined as:

– If a weight is less than -0.01, it is set to -0.01.

– If a weight is more than 0.01, it is set to 0.01.

I’m not sure it is an appropriate loss function for a normal image classification task. i.e. you cannot.

Thank you very much, Sir

You’re welcome.

Great article!

I trained the GAN on all 9 digits and the results are wild. Lots of digits that look halfway between one digit and another… my model loves generating 0&8 hybrids. Have a few ideas for limiting this effect such as : using a bigger latent space, feeding the class as input into the network, using a bigger conv model, maybe adding a fc layer that appends to the output of the discriminator that my model can use as a class predictor(??) and obviously training longer.

Well done James!

So i implemented your model just as you provided it, but my loss functions look entirely different. Crit_real and crit_fake drop all the way down to -150 after 400 epochs while gen goes to 150 after 400 epochs.

Up to that point the resulting images look good, but afterwards they just turn all black.

Can you give an explanation why this would happen?

You changed the loss function and got bad results. Sorry, I don’t know why other than you changed the loss function.

Perhaps try the code as listed first?

I used your model, once local on a MS Surface Pro 3 (very slow of course) and once with Colab from Google, got very different results for the coefficiants and also for the final plot of loss and accuracy. I changed nothing in the proposed loss function.

Can this be an expected result

Yes, sometimes the models can be quite unstable, they can be tricky.

I changed the machine in Colab from CPU to GNU and the results are similar to the script

Nice!

I am trying to use Wasserstein loss for pix2pix and yet haven’t got any good results. one strange thing that happens is that the last hundred steps, it doesn’t matter if I have 1000 steps or 5000 steps, the last hundred steps suddenly the g_loss starts spiking, for example

>935, c1=-8.903, c2=-5.745 g=9.406

>936, c1=-12.613, c2=-9.001 g=-5.578

>937, c1=-9.143, c2=-8.499 g=-8.585

>938, c1=-12.172, c2=-5.203 g=1.296

>939, c1=-11.419, c2=-7.457 g=2.835

>940, c1=-11.785, c2=-7.998 g=-6.124

>941, c1=-12.427, c2=-8.275 g=-2.679

>942, c1=-12.758, c2=-8.763 g=4.127

>943, c1=-9.551, c2=-6.013 g=6.745

>944, c1=-9.443, c2=-8.791 g=-3.048

>945, c1=-10.753, c2=-7.275 g=-2.918

>946, c1=-12.762, c2=-8.732 g=10.906

>947, c1=-10.392, c2=-5.713 g=-7.287

>948, c1=-12.502, c2=-8.810 g=0.580

>949, c1=-12.936, c2=-9.329 g=-2.742

>950, c1=-7.656, c2=-7.766 g=-1.436

>951, c1=-10.732, c2=-4.744 g=-7.908

>952, c1=-12.870, c2=-9.123 g=5.914

>953, c1=-9.666, c2=-7.812 g=6.430

>954, c1=-12.867, c2=-7.669 g=11.883

>955, c1=-10.069, c2=-9.066 g=-4.029

>956, c1=-13.045, c2=-9.065 g=11.350

>957, c1=-11.683, c2=-0.542 g=11.850

>958, c1=-10.341, c2=-7.843 g=-0.476

>959, c1=-13.147, c2=-9.353 g=-3.374

>960, c1=-6.914, c2=-6.306 g=-3.759

even if I reduce the steps to 500, still at the last steps I get this.

another problem is my g-loss starts at like 80 and then comes down and it doesn’t start as 0.

I don’t get good results at all. Do you have any suggestions?

Good question.

Hmmm, I don’t have any good off the cuff suggestions other than ensure the implementation is correct and try making small changes to see if you can uncover the cause of the fault. Debug!

Let me know how you go.

Hi,

Thanks for the post. It was really helpful. I am planning to do a project about generating fake sentences/text/reviews based Text data. I did some research on online where I found out softmax encoder/decoder is the best way generate fake text for GAN. Another way is Reinforcement Learning though. Can you give me some ideas about how I can use Text data instead of image data?

I would recommend a language model for text instead of a GAN:

https://machinelearningmastery.com/start-here/#nlp

Thanks for the all the links for NLP. Can we break the words as vectors and feed them to discriminators and then generate some random text to figure out whether the reviews/sentences are actually from data or just generated from Generator?

I don’t see why not.

Hi Jason,

I am trying load dataset from the directory which has some .jpg images. I am trying to follow your load datatset tutorial but it seems there is some issues. When I ran the code the program goes like forever. Could you please guide something for loading custom dataset image like cat, cars instead of built in mnist dataset? Here is my code and output.

trainX = datagen.flow_from_directory(‘/celeb/train’, class_mode=’binary’, batch_size=64)

X = np.array(trainX, dtype=’float32′)

X = expand_dims(X, axis=-1)

X = X.astype(‘float32’)

X = (X – 127.5) / 127.5

print(trainX.shape)

Found 93 images belonging to 5 classes.

Program got stuck. No result for printing X or trainX value.

Perhaps this will help:

https://machinelearningmastery.com/how-to-load-large-datasets-from-directories-for-deep-learning-with-keras/

For some reason, when I try to run this example using Keras from Tensorflow 2, it doesn’t converge.

I’ve tried it both in Colab and Kaggle and it only worked when I downgraded my TF version to 1.14.

Did anyone else had this issue?

GANs do not converge, they find equilibrium.

Perhaps start with some of the simpler tutorials here to understand GAN basics:

https://machinelearningmastery.com/start-here/#gans

In the complete example, on line 209 of the train() method why do we have “y_gan = -ones((n_batch, 1))”? Wouldn’t this give the fake samples a label of -1? I thought that real samples have a label of -1 while the fake samples have a label of 1.

Ouch, I should have been consistent, sorry.

Both approaches work the same.

In the complete example on line 58, why is the kernel_constraint parameter not provided for the Dense layer? Is weight clipping not required in this layer of the critic, and if so, why?

Thank you for this awesome tutorial, by the way!

That is the output layer, I typically don’t add constraints to output layers.

Why? Habit/experience I guess. Perhaps try it and see.

Any tips on adapting this for using a gradient penalty instead of weight clipping?

Your series on GANs is currently the most helpful on the internet IMO, thanks!

Thanks.

No. From memory, the gradient penalty was a challenge to implement in Keras. I’m sure some bright mind figured it out though.

Hi Jason

Read one of you books… mailed a few times… 🙂

I think this is what Lennert S’s problem was. Because mine is similar. On Tensor flow 1.15 this stabalises happily and i get good 7’s for both CPU and GPU. Copy paste of your code.

On Tensorflow 2.0+ I get 100 little black squares and It does not stabilise. There is definately a difference here. I previously narrowed it down to the batchnorm when i was playing with GANs and removing BN improved things a bit. It doesn’t for WGAN so i’m not sure.

Wondering if you iunderstand whats going on here? Have you tried this with TF 2.0?

I can mail you my (your) plots if you like?

Greg

Fascinating. I have not seen this myself. I will investigate (adding to trello now).

UPDATE: I do not see any problem with Keras 2.3 and Tensoflow 2.1.

Try running the example a few times and try inspecting the results from different epochs.

Some suggestions here:

https://machinelearningmastery.com/faq/single-faq/why-does-the-code-in-the-tutorial-not-work-for-me

Hi Jason,

I guess, you do mean “iterations” (steps) not epochs, in your loss graphs along x-axis? There are 10 epochs, each having 97 steps. Otherwise, I obtain the results for “7”, similar to yours. I observe quite a sharp “transition” at about step 194 when all generated images turn extremely dark, but after one epoch this darkness fades away, primarily from the background, thus leaving a beautiful “7” image alone. Thanks for the tutorial!

You’re welcome!

I believe, I get it now: in the first step (inner loop) you are trying to find a maximal difference between both fixed statistical distributions of the real and generated sets, by adjusting w-parameters of the critic and their subsequent clipping. This is needed to ensure, that the difference stays closer to the true Wasserstein distance, i.e. that we are in a Lipshitz space . In the second step (oughter loop) you are trying to minimize this best-defined difference, by adjusting the generator (via theta-parameters) to bring its distribution closer to the real one.

What I don’t understand – is why do c2 and g losses have different signs? In both cases it is the same generator loss estimated by the critic. In both cases it is evaluated on the fake samples and printed out as it is.

We are trying to improve the generator (g), which is the inverse of the capability/expectation of the critic (c).

Hi, I am confused of the loss function, where two losses of generator and discriminator presented in the paper are different, so how do you transform the two losses into the same loss function in this course. Thank you, Jason.

Perhaps this tutorial will help:

https://machinelearningmastery.com/how-to-implement-wasserstein-loss-for-generative-adversarial-networks/

Hi Jason,

Do you have any implementations regarding the Improved WGAN method?

Not at this stage.