Watch keynotes, product sessions, workshops, and more from Google I/O See playlist


View on Run in Google Colab View source on GitHub Download notebook

This notebook demonstrates image to image translation using conditional GAN's, as described in Image-to-Image Translation with Conditional Adversarial Networks. Using this technique you can colorize black and white photos, convert google maps to google earth, etc. Here, convert building facades to real buildings.

In example, you will use the CMP Facade Database, helpfully provided by the Center for Machine Perception at the Czech Technical University in Prague. To keep the example short, use a preprocessed copy of this dataset, created by the authors of the paper above.

Each epoch takes around 15 seconds on a single V100 GPU.

Below is the output generated after training the model for 200 epochs.

sample output_1 sample output_2

Import TensorFlow and other libraries

import tensorflow as tf

import os
import time

from matplotlib import pyplot as plt
from IPython import display
pip install -q -U tensorboard

Load the dataset

You can download this dataset and similar datasets from here. As mentioned in the paper, apply random jittering and mirroring to the training dataset.

  • In random jittering, the image is resized to 286 x 286 and then randomly cropped to 256 x 256
  • In random mirroring, the image is randomly flipped horizontally i.e left to right.
_URL = ''

path_to_zip = tf.keras.utils.get_file('facades.tar.gz',

PATH = os.path.join(os.path.dirname(path_to_zip), 'facades/')
Downloading data from
30171136/30168306 [==============================] - 10s 0us/step
def load(image_file):
  image =
  image = tf.image.decode_jpeg(image)

  w = tf.shape(image)[1]

  w = w // 2
  real_image = image[:, :w, :]
  input_image = image[:, w:, :]

  input_image = tf.cast(input_image, tf.float32)
  real_image = tf.cast(real_image, tf.float32)

  return input_image, real_image
inp, re = load(PATH+'train/100.jpg')
# casting to int for matplotlib to show the image
<matplotlib.image.AxesImage at 0x7f194823a7d0>



def resize(input_image, real_image, height, width):
  input_image = tf.image.resize(input_image, [height, width],
  real_image = tf.image.resize(real_image, [height, width],

  return input_image, real_image
def random_crop(input_image, real_image):
  stacked_image = tf.stack([input_image, real_image], axis=0)
  cropped_image = tf.image.random_crop(
      stacked_image, size=[2, IMG_HEIGHT, IMG_WIDTH, 3])

  return cropped_image[0], cropped_image[1]
# normalizing the images to [-1, 1]

def normalize(input_image, real_image):
  input_image = (input_image / 127.5) - 1
  real_image = (real_image / 127.5) - 1

  return input_image, real_image
def random_jitter(input_image, real_image):
  # resizing to 286 x 286 x 3
  input_image, real_image = resize(input_image, real_image, 286, 286)

  # randomly cropping to 256 x 256 x 3
  input_image, real_image = random_crop(input_image, real_image)

  if tf.random.uniform(()) > 0.5:
    # random mirroring
    input_image = tf.image.flip_left_right(input_image)
    real_image = tf.image.flip_left_right(real_image)

  return input_image, real_image

As you can see in the images below that they are going through random jittering Random jittering as described in the paper is to

  1. Resize an image to bigger height and width
  2. Randomly crop to the target size
  3. Randomly flip the image horizontally
plt.figure(figsize=(6, 6))
for i in range(4):
  rj_inp, rj_re = random_jitter(inp, re)
  plt.subplot(2, 2, i+1)


def load_image_train(image_file):
  input_image, real_image = load(image_file)
  input_image, real_image = random_jitter(input_image, real_image)
  input_image, real_image = normalize(input_image, real_image)

  return input_image, real_image
def load_image_test(image_file):
  input_image, real_image = load(image_file)
  input_image, real_image = resize(input_image, real_image,
                                   IMG_HEIGHT, IMG_WIDTH)
  input_image, real_image = normalize(input_image, real_image)

  return input_image, real_image

Input Pipeline

train_dataset ='train/*.jpg')
train_dataset =,
train_dataset = train_dataset.shuffle(BUFFER_SIZE)
train_dataset = train_dataset.batch(BATCH_SIZE)
test_dataset ='test/*.jpg')
test_dataset =
test_dataset = test_dataset.batch(BATCH_SIZE)

Build the Generator

  • The architecture of generator is a modified U-Net.
  • Each block in the encoder is (Conv -> Batchnorm -> Leaky ReLU)
  • Each block in the decoder is (Transposed Conv -> Batchnorm -> Dropout(applied to the first 3 blocks) -> ReLU)
  • There are skip connections between the encoder and decoder (as in U-Net).
def downsample(filters, size, apply_batchnorm=True):
  initializer = tf.random_normal_initializer(0., 0.02)

  result = tf.keras.Sequential()
      tf.keras.layers.Conv2D(filters, size, strides=2, padding='same',
                             kernel_initializer=initializer, use_bias=False))

  if apply_batchnorm:


  return result
down_model = downsample(3, 4)
down_result = down_model(tf.expand_dims(inp, 0))
print (down_result.shape)
(1, 128, 128, 3)
def upsample(filters, size, apply_dropout=False):
  initializer = tf.random_normal_initializer(0., 0.02)

  result = tf.keras.Sequential()
    tf.keras.layers.Conv2DTranspose(filters, size, strides=2,


  if apply_dropout:


  return result
up_model = upsample(3, 4)
up_result = up_model(down_result)
print (up_result.shape)
(1, 256, 256, 3)
def Generator():
  inputs = tf.keras.layers.Input(shape=[256, 256, 3])

  down_stack = [
    downsample(64, 4, apply_batchnorm=False),  # (bs, 128, 128, 64)
    downsample(128, 4),  # (bs, 64, 64, 128)
    downsample(256, 4),  # (bs, 32, 32, 256)
    downsample(512, 4),  # (bs, 16, 16, 512)
    downsample(512, 4),  # (bs, 8, 8, 512)
    downsample(512, 4),  # (bs, 4, 4, 512)
    downsample(512, 4),  # (bs, 2, 2, 512)
    downsample(512, 4),  # (bs, 1, 1, 512)

  up_stack = [
    upsample(512, 4, apply_dropout=True),  # (bs, 2, 2, 1024)
    upsample(512, 4, apply_dropout=True),  # (bs, 4, 4, 1024)
    upsample(512, 4, apply_dropout=True),  # (bs, 8, 8, 1024)
    upsample(512, 4),  # (bs, 16, 16, 1024)
    upsample(256, 4),  # (bs, 32, 32, 512)
    upsample(128, 4),  # (bs, 64, 64, 256)
    upsample(64, 4),  # (bs, 128, 128, 128)

  initializer = tf.random_normal_initializer(0., 0.02)
  last = tf.keras.layers.Conv2DTranspose(OUTPUT_CHANNELS, 4,
                                         activation='tanh')  # (bs, 256, 256, 3)

  x = inputs

  # Downsampling through the model
  skips = []
  for down in down_stack:
    x = down(x)

  skips = reversed(skips[:-1])

  # Upsampling and establishing the skip connections
  for up, skip in zip(up_stack, skips):
    x = up(x)
    x = tf.keras.layers.Concatenate()([x, skip])

  x = last(x)

  return tf.keras.Model(inputs=inputs, outputs=x)
generator = Generator()
tf.keras.utils.plot_model(generator, show_shapes=True, dpi=64)


gen_output = generator(inp[tf.newaxis, ...], training=False)
plt.imshow(gen_output[0, ...])
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
<matplotlib.image.AxesImage at 0x7f189c184e10>


  • Generator loss
    • It is a sigmoid cross entropy loss of the generated images and an array of ones.
    • The paper also includes L1 loss which is MAE (mean absolute error) between the generated image and the target image.
    • This allows the generated image to become structurally similar to the target image.
    • The formula to calculate the total generator loss = gan_loss + LAMBDA * l1_loss, where LAMBDA = 100. This value was decided by the authors of the paper.

The training procedure for the generator is shown below:

LAMBDA = 100
loss_object = tf.keras.losses.BinaryCrossentropy(from_logits=True)
def generator_loss(disc_generated_output, gen_output, target):
  gan_loss = loss_object(tf.ones_like(disc_generated_output), disc_generated_output)

  # mean absolute error
  l1_loss = tf.reduce_mean(tf.abs(target - gen_output))

  total_gen_loss = gan_loss + (LAMBDA * l1_loss)

  return total_gen_loss, gan_loss, l1_loss

Generator Update Image

Build the Discriminator

  • The Discriminator is a PatchGAN.
  • Each block in the discriminator is (Conv -> BatchNorm -> Leaky ReLU)
  • The shape of the output after the last layer is (batch_size, 30, 30, 1)
  • Each 30x30 patch of the output classifies a 70x70 portion of the input image (such an architecture is called a PatchGAN).
  • Discriminator receives 2 inputs.
    • Input image and the target image, which it should classify as real.
    • Input image and the generated image (output of generator), which it should classify as fake.
    • Concatenate these 2 inputs together in the code (tf.concat([inp, tar], axis=-1))
def Discriminator():
  initializer = tf.random_normal_initializer(0., 0.02)

  inp = tf.keras.layers.Input(shape=[256, 256, 3], name='input_image')
  tar = tf.keras.layers.Input(shape=[256, 256, 3], name='target_image')

  x = tf.keras.layers.concatenate([inp, tar])  # (bs, 256, 256, channels*2)

  down1 = downsample(64, 4, False)(x)  # (bs, 128, 128, 64)
  down2 = downsample(128, 4)(down1)  # (bs, 64, 64, 128)
  down3 = downsample(256, 4)(down2)  # (bs, 32, 32, 256)

  zero_pad1 = tf.keras.layers.ZeroPadding2D()(down3)  # (bs, 34, 34, 256)
  conv = tf.keras.layers.Conv2D(512, 4, strides=1,
                                use_bias=False)(zero_pad1)  # (bs, 31, 31, 512)

  batchnorm1 = tf.keras.layers.BatchNormalization()(conv)

  leaky_relu = tf.keras.layers.LeakyReLU()(batchnorm1)

  zero_pad2 = tf.keras.layers.ZeroPadding2D()(leaky_relu)  # (bs, 33, 33, 512)

  last = tf.keras.layers.Conv2D(1, 4, strides=1,
                                kernel_initializer=initializer)(zero_pad2)  # (bs, 30, 30, 1)

  return tf.keras.Model(inputs=[inp, tar], outputs=last)
discriminator = Discriminator()
tf.keras.utils.plot_model(discriminator, show_shapes=True, dpi=64)


disc_out = discriminator([inp[tf.newaxis, ...], gen_output], training=False)
plt.imshow(disc_out[0, ..., -1], vmin=-20, vmax=20, cmap='RdBu_r')
<matplotlib.colorbar.Colorbar at 0x7f183c3cd550>


Discriminator loss

  • The discriminator loss function takes 2 inputs; real images, generated images
  • real_loss is a sigmoid cross entropy loss of the real images and an array of ones(since these are the real images)
  • generated_loss is a sigmoid cross entropy loss of the generated images and an array of zeros(since these are the fake images)
  • Then the total_loss is the sum of real_loss and the generated_loss
def discriminator_loss(disc_real_output, disc_generated_output):
  real_loss = loss_object(tf.ones_like(disc_real_output), disc_real_output)

  generated_loss = loss_object(tf.zeros_like(disc_generated_output), disc_generated_output)

  total_disc_loss = real_loss + generated_loss

  return total_disc_loss

The training procedure for the discriminator is shown below.

To learn more about the architecture and the hyperparameters you can refer the paper.

Discriminator Update Image

Define the Optimizers and Checkpoint-saver

generator_optimizer = tf.keras.optimizers.Adam(2e-4, beta_1=0.5)
discriminator_optimizer = tf.keras.optimizers.Adam(2e-4, beta_1=0.5)
checkpoint_dir = './training_checkpoints'
checkpoint_prefix = os.path.join(checkpoint_dir, "ckpt")
checkpoint = tf.train.Checkpoint(generator_optimizer=generator_optimizer,

Generate Images

Write a function to plot some images during training.

  • Pass images from the test dataset to the generator.
  • The generator will then translate the input image into the output.
  • Last step is to plot the predictions and voila!
def generate_images(model, test_input, tar):
  prediction = model(test_input, training=True)
  plt.figure(figsize=(15, 15))

  display_list = [test_input[0], tar[0], prediction[0]]
  title = ['Input Image', 'Ground Truth', 'Predicted Image']

  for i in range(3):
    plt.subplot(1, 3, i+1)
    # getting the pixel values between [0, 1] to plot it.
    plt.imshow(display_list[i] * 0.5 + 0.5)
for example_input, example_target in test_dataset.take(1):
  generate_images(generator, example_input, example_target)



  • For each example input generate an output.
  • The discriminator receives the input_image and the generated image as the first input. The second input is the input_image and the target_image.
  • Next, calculate the generator and the discriminator loss.
  • Then, calculate the gradients of loss with respect to both the generator and the discriminator variables(inputs) and apply those to the optimizer.
  • Then log the losses to TensorBoard.
EPOCHS = 150
import datetime

summary_writer = tf.summary.create_file_writer(
  log_dir + "fit/" +"%Y%m%d-%H%M%S"))
def train_step(input_image, target, epoch):
  with tf.GradientTape() as gen_tape, tf.GradientTape() as disc_tape:
    gen_output = generator(input_image, training=True)

    disc_real_output = discriminator([input_image, target], training=True)
    disc_generated_output = discriminator([input_image, gen_output], training=True)

    gen_total_loss, gen_gan_loss, gen_l1_loss = generator_loss(disc_generated_output, gen_output, target)
    disc_loss = discriminator_loss(disc_real_output, disc_generated_output)

  generator_gradients = gen_tape.gradient(gen_total_loss,
  discriminator_gradients = disc_tape.gradient(disc_loss,


  with summary_writer.as_default():
    tf.summary.scalar('gen_total_loss', gen_total_loss, step=epoch)
    tf.summary.scalar('gen_gan_loss', gen_gan_loss, step=epoch)
    tf.summary.scalar('gen_l1_loss', gen_l1_loss, step=epoch)
    tf.summary.scalar('disc_loss', disc_loss, step=epoch)

The actual training loop:

  • Iterates over the number of epochs.
  • On each epoch it clears the display, and runs generate_images to show it's progress.
  • On each epoch it iterates over the training dataset, printing a '.' for each example.
  • It saves a checkpoint every 20 epochs.
def fit(train_ds, epochs, test_ds):
  for epoch in range(epochs):
    start = time.time()


    for example_input, example_target in test_ds.take(1):
      generate_images(generator, example_input, example_target)
    print("Epoch: ", epoch)

    # Train
    for n, (input_image, target) in train_ds.enumerate():
      print('.', end='')
      if (n+1) % 100 == 0:
      train_step(input_image, target, epoch)

    # saving (checkpoint) the model every 20 epochs
    if (epoch + 1) % 20 == 0:

    print ('Time taken for epoch {} is {} sec\n'.format(epoch + 1,

This training loop saves logs you can easily view in TensorBoard to monitor the training progress. Working locally you would launch a separate tensorboard process. In a notebook, if you want to monitor with TensorBoard it's easiest to launch the viewer before starting the training.

To launch the viewer paste the following into a code-cell:

%load_ext tensorboard
%tensorboard --logdir {log_dir}

Now run the training loop:

fit(train_dataset, EPOCHS, test_dataset)


Epoch:  149

Time taken for epoch 150 is 15.598696947097778 sec

If you want to share the TensorBoard results publicly you can upload the logs to by copying the following into a code-cell.

tensorboard dev upload --logdir  {log_dir}

You can view the results of a previous run of this notebook on is a managed experience for hosting, tracking, and sharing ML experiments with everyone.

It can also included inline using an <iframe>:


Interpreting the logs from a GAN is more subtle than a simple classification or regression model. Things to look for::

  • Check that neither model has "won". If either the gen_gan_loss or the disc_loss gets very low it's an indicator that this model is dominating the other, and you are not successfully training the combined model.
  • The value log(2) = 0.69 is a good reference point for these losses, as it indicates a perplexity of 2: That the discriminator is on average equally uncertain about the two options.
  • For the disc_loss a value below 0.69 means the discriminator is doing better than random, on the combined set of real+generated images.
  • For the gen_gan_loss a value below 0.69 means the generator is doing better than random at fooling the descriminator.
  • As training progresses the gen_l1_loss should go down.

Restore the latest checkpoint and test

ls {checkpoint_dir}
checkpoint  ckpt-5.index
ckpt-1.index    ckpt-6.index
ckpt-2.index    ckpt-7.index
ckpt-3.index    ckpt-8.index
# restoring the latest checkpoint in checkpoint_dir
< at 0x7f134de35a90>

Generate using test dataset

# Run the trained model on a few examples from the test dataset
for inp, tar in test_dataset.take(5):
  generate_images(generator, inp, tar)