Deep Convolutional Generative Adversarial Network

View on TensorFlow.org Run in Google Colab View source on GitHub Download notebook

This tutorial demonstrates how to generate images of handwritten digits using a Deep Convolutional Generative Adversarial Network (DCGAN). The code is written using the Keras Sequential API with a tf.GradientTape training loop.

What are GANs?

Generative Adversarial Networks (GANs) are one of the most interesting ideas in computer science today. Two models are trained simultaneously by an adversarial process. A generator ("the artist") learns to create images that look real, while a discriminator ("the art critic") learns to tell real images apart from fakes.

A diagram of a generator and discriminator

During training, the generator progressively becomes better at creating images that look real, while the discriminator becomes better at telling them apart. The process reaches equilibrium when the discriminator can no longer distinguish real images from fakes.

A second diagram of a generator and discriminator

This notebook demonstrates this process on the MNIST dataset. The following animation shows a series of images produced by the generator as it was trained for 50 epochs. The images begin as random noise, and increasingly resemble hand written digits over time.

sample output

To learn more about GANs, see MIT's Intro to Deep Learning course.

Setup

import tensorflow as tf
2023-11-05 04:03:54.774427: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
2023-11-05 04:03:54.774476: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
2023-11-05 04:03:54.775979: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
tf.__version__
'2.15.0-rc1'
# To generate GIFs
pip install imageio
pip install git+https://github.com/tensorflow/docs
import glob
import imageio
import matplotlib.pyplot as plt
import numpy as np
import os
import PIL
from tensorflow.keras import layers
import time

from IPython import display

Load and prepare the dataset

You will use the MNIST dataset to train the generator and the discriminator. The generator will generate handwritten digits resembling the MNIST data.

(train_images, train_labels), (_, _) = tf.keras.datasets.mnist.load_data()
train_images = train_images.reshape(train_images.shape[0], 28, 28, 1).astype('float32')
train_images = (train_images - 127.5) / 127.5  # Normalize the images to [-1, 1]
BUFFER_SIZE = 60000
BATCH_SIZE = 256
# Batch and shuffle the data
train_dataset = tf.data.Dataset.from_tensor_slices(train_images).shuffle(BUFFER_SIZE).batch(BATCH_SIZE)

Create the models

Both the generator and discriminator are defined using the Keras Sequential API.

The Generator

The generator uses tf.keras.layers.Conv2DTranspose (upsampling) layers to produce an image from a seed (random noise). Start with a Dense layer that takes this seed as input, then upsample several times until you reach the desired image size of 28x28x1. Notice the tf.keras.layers.LeakyReLU activation for each layer, except the output layer which uses tanh.

def make_generator_model():
    model = tf.keras.Sequential()
    model.add(layers.Dense(7*7*256, use_bias=False, input_shape=(100,)))
    model.add(layers.BatchNormalization())
    model.add(layers.LeakyReLU())

    model.add(layers.Reshape((7, 7, 256)))
    assert model.output_shape == (None, 7, 7, 256)  # Note: None is the batch size

    model.add(layers.Conv2DTranspose(128, (5, 5), strides=(1, 1), padding='same', use_bias=False))
    assert model.output_shape == (None, 7, 7, 128)
    model.add(layers.BatchNormalization())
    model.add(layers.LeakyReLU())

    model.add(layers.Conv2DTranspose(64, (5, 5), strides=(2, 2), padding='same', use_bias=False))
    assert model.output_shape == (None, 14, 14, 64)
    model.add(layers.BatchNormalization())
    model.add(layers.LeakyReLU())

    model.add(layers.Conv2DTranspose(1, (5, 5), strides=(2, 2), padding='same', use_bias=False, activation='tanh'))
    assert model.output_shape == (None, 28, 28, 1)

    return model

Use the (as yet untrained) generator to create an image.

generator = make_generator_model()

noise = tf.random.normal([1, 100])
generated_image = generator(noise, training=False)

plt.imshow(generated_image[0, :, :, 0], cmap='gray')
<matplotlib.image.AxesImage at 0x7fa106d23a00>

png

The Discriminator

The discriminator is a CNN-based image classifier.

def make_discriminator_model():
    model = tf.keras.Sequential()
    model.add(layers.Conv2D(64, (5, 5), strides=(2, 2), padding='same',
                                     input_shape=[28, 28, 1]))
    model.add(layers.LeakyReLU())
    model.add(layers.Dropout(0.3))

    model.add(layers.Conv2D(128, (5, 5), strides=(2, 2), padding='same'))
    model.add(layers.LeakyReLU())
    model.add(layers.Dropout(0.3))

    model.add(layers.Flatten())
    model.add(layers.Dense(1))

    return model

Use the (as yet untrained) discriminator to classify the generated images as real or fake. The model will be trained to output positive values for real images, and negative values for fake images.

discriminator = make_discriminator_model()
decision = discriminator(generated_image)
print (decision)
tf.Tensor([[-0.00071298]], shape=(1, 1), dtype=float32)

Define the loss and optimizers

Define loss functions and optimizers for both models.

# This method returns a helper function to compute cross entropy loss
cross_entropy = tf.keras.losses.BinaryCrossentropy(from_logits=True)

Discriminator loss

This method quantifies how well the discriminator is able to distinguish real images from fakes. It compares the discriminator's predictions on real images to an array of 1s, and the discriminator's predictions on fake (generated) images to an array of 0s.

def discriminator_loss(real_output, fake_output):
    real_loss = cross_entropy(tf.ones_like(real_output), real_output)
    fake_loss = cross_entropy(tf.zeros_like(fake_output), fake_output)
    total_loss = real_loss + fake_loss
    return total_loss

Generator loss

The generator's loss quantifies how well it was able to trick the discriminator. Intuitively, if the generator is performing well, the discriminator will classify the fake images as real (or 1). Here, compare the discriminators decisions on the generated images to an array of 1s.

def generator_loss(fake_output):
    return cross_entropy(tf.ones_like(fake_output), fake_output)

The discriminator and the generator optimizers are different since you will train two networks separately.

generator_optimizer = tf.keras.optimizers.Adam(1e-4)
discriminator_optimizer = tf.keras.optimizers.Adam(1e-4)

Save checkpoints

This notebook also demonstrates how to save and restore models, which can be helpful in case a long running training task is interrupted.

checkpoint_dir = './training_checkpoints'
checkpoint_prefix = os.path.join(checkpoint_dir, "ckpt")
checkpoint = tf.train.Checkpoint(generator_optimizer=generator_optimizer,
                                 discriminator_optimizer=discriminator_optimizer,
                                 generator=generator,
                                 discriminator=discriminator)

Define the training loop

EPOCHS = 50
noise_dim = 100
num_examples_to_generate = 16

# You will reuse this seed overtime (so it's easier)
# to visualize progress in the animated GIF)
seed = tf.random.normal([num_examples_to_generate, noise_dim])

The training loop begins with generator receiving a random seed as input. That seed is used to produce an image. The discriminator is then used to classify real images (drawn from the training set) and fakes images (produced by the generator). The loss is calculated for each of these models, and the gradients are used to update the generator and discriminator.

# Notice the use of `tf.function`
# This annotation causes the function to be "compiled".
@tf.function
def train_step(images):
    noise = tf.random.normal([BATCH_SIZE, noise_dim])

    with tf.GradientTape() as gen_tape, tf.GradientTape() as disc_tape:
      generated_images = generator(noise, training=True)

      real_output = discriminator(images, training=True)
      fake_output = discriminator(generated_images, training=True)

      gen_loss = generator_loss(fake_output)
      disc_loss = discriminator_loss(real_output, fake_output)

    gradients_of_generator = gen_tape.gradient(gen_loss, generator.trainable_variables)
    gradients_of_discriminator = disc_tape.gradient(disc_loss, discriminator.trainable_variables)

    generator_optimizer.apply_gradients(zip(gradients_of_generator, generator.trainable_variables))
    discriminator_optimizer.apply_gradients(zip(gradients_of_discriminator, discriminator.trainable_variables))
def train(dataset, epochs):
  for epoch in range(epochs):
    start = time.time()

    for image_batch in dataset:
      train_step(image_batch)

    # Produce images for the GIF as you go
    display.clear_output(wait=True)
    generate_and_save_images(generator,
                             epoch + 1,
                             seed)

    # Save the model every 15 epochs
    if (epoch + 1) % 15 == 0:
      checkpoint.save(file_prefix = checkpoint_prefix)

    print ('Time for epoch {} is {} sec'.format(epoch + 1, time.time()-start))

  # Generate after the final epoch
  display.clear_output(wait=True)
  generate_and_save_images(generator,
                           epochs,
                           seed)

Generate and save images

def generate_and_save_images(model, epoch, test_input):
  # Notice `training` is set to False.
  # This is so all layers run in inference mode (batchnorm).
  predictions = model(test_input, training=False)

  fig = plt.figure(figsize=(4, 4))

  for i in range(predictions.shape[0]):
      plt.subplot(4, 4, i+1)
      plt.imshow(predictions[i, :, :, 0] * 127.5 + 127.5, cmap='gray')
      plt.axis('off')

  plt.savefig('image_at_epoch_{:04d}.png'.format(epoch))
  plt.show()

Train the model

Call the train() method defined above to train the generator and discriminator simultaneously. Note, training GANs can be tricky. It's important that the generator and discriminator do not overpower each other (e.g., that they train at a similar rate).

At the beginning of the training, the generated images look like random noise. As training progresses, the generated digits will look increasingly real. After about 50 epochs, they resemble MNIST digits. This may take about one minute / epoch with the default settings on Colab.

train(train_dataset, EPOCHS)

png

Restore the latest checkpoint.

checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))
<tensorflow.python.checkpoint.checkpoint.CheckpointLoadStatus at 0x7fa106bd6eb0>

Create a GIF

# Display a single image using the epoch number
def display_image(epoch_no):
  return PIL.Image.open('image_at_epoch_{:04d}.png'.format(epoch_no))
display_image(EPOCHS)

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

KeyError                                  Traceback (most recent call last)

File /tmpfs/src/tf_docs_env/lib/python3.9/site-packages/PIL/JpegImagePlugin.py:639, in _save(im, fp, filename)
    638 try:
--> 639     rawmode = RAWMODE[im.mode]
    640 except KeyError as e:


KeyError: 'RGBA'


The above exception was the direct cause of the following exception:


OSError                                   Traceback (most recent call last)

File /tmpfs/src/tf_docs_env/lib/python3.9/site-packages/PIL/Image.py:643, in Image._repr_image(self, image_format, **kwargs)
    642 try:
--> 643     self.save(b, image_format, **kwargs)
    644 except Exception as e:


File /tmpfs/src/tf_docs_env/lib/python3.9/site-packages/PIL/Image.py:2413, in Image.save(self, fp, format, **params)
   2412 try:
-> 2413     save_handler(self, fp, filename)
   2414 except Exception:


File /tmpfs/src/tf_docs_env/lib/python3.9/site-packages/PIL/JpegImagePlugin.py:642, in _save(im, fp, filename)
    641     msg = f"cannot write mode {im.mode} as JPEG"
--> 642     raise OSError(msg) from e
    644 info = im.encoderinfo


OSError: cannot write mode RGBA as JPEG


The above exception was the direct cause of the following exception:


ValueError                                Traceback (most recent call last)

File /tmpfs/src/tf_docs_env/lib/python3.9/site-packages/IPython/core/formatters.py:344, in BaseFormatter.__call__(self, obj)
    342     method = get_real_method(obj, self.print_method)
    343     if method is not None:
--> 344         return method()
    345     return None
    346 else:


File /tmpfs/src/tf_docs_env/lib/python3.9/site-packages/PIL/Image.py:661, in Image._repr_jpeg_(self)
    656 def _repr_jpeg_(self):
    657     """iPython display hook support for JPEG format.
    658 
    659     :returns: JPEG version of the image as bytes
    660     """
--> 661     return self._repr_image("JPEG")


File /tmpfs/src/tf_docs_env/lib/python3.9/site-packages/PIL/Image.py:646, in Image._repr_image(self, image_format, **kwargs)
    644 except Exception as e:
    645     msg = f"Could not save to {image_format} for display"
--> 646     raise ValueError(msg) from e
    647 return b.getvalue()


ValueError: Could not save to JPEG for display

png

Use imageio to create an animated gif using the images saved during training.

anim_file = 'dcgan.gif'

with imageio.get_writer(anim_file, mode='I') as writer:
  filenames = glob.glob('image*.png')
  filenames = sorted(filenames)
  for filename in filenames:
    image = imageio.imread(filename)
    writer.append_data(image)
  image = imageio.imread(filename)
  writer.append_data(image)
/tmpfs/tmp/ipykernel_42443/1982054950.py:7: DeprecationWarning: Starting with ImageIO v3 the behavior of this function will switch to that of iio.v3.imread. To keep the current behavior (and make this warning disappear) use `import imageio.v2 as imageio` or call `imageio.v2.imread` directly.
  image = imageio.imread(filename)
/tmpfs/tmp/ipykernel_42443/1982054950.py:9: DeprecationWarning: Starting with ImageIO v3 the behavior of this function will switch to that of iio.v3.imread. To keep the current behavior (and make this warning disappear) use `import imageio.v2 as imageio` or call `imageio.v2.imread` directly.
  image = imageio.imread(filename)
import tensorflow_docs.vis.embed as embed
embed.embed_file(anim_file)

gif

Next steps

This tutorial has shown the complete code necessary to write and train a GAN. As a next step, you might like to experiment with a different dataset, for example the Large-scale Celeb Faces Attributes (CelebA) dataset available on Kaggle. To learn more about GANs see the NIPS 2016 Tutorial: Generative Adversarial Networks.