Producing photos with Keras and TensorFlow keen execution

The latest announcement of TensorFlow 2.0 names keen execution because the primary central characteristic of the brand new main model. What does this imply for R customers?
As demonstrated in our latest submit on neural machine translation, you need to use keen execution from R now already, together with Keras customized fashions and the datasets API. It’s good to know you can use it – however why do you have to? And by which circumstances?

On this and some upcoming posts, we wish to present how keen execution could make creating fashions lots simpler. The diploma of simplication will depend upon the duty – and simply how a lot simpler you’ll discover the brand new approach may also rely in your expertise utilizing the useful API to mannequin extra advanced relationships.
Even if you happen to suppose that GANs, encoder-decoder architectures, or neural fashion switch didn’t pose any issues earlier than the appearance of keen execution, you may discover that the choice is a greater match to how we people mentally image issues.

For this submit, we’re porting code from a latest Google Colaboratory pocket book implementing the DCGAN structure.(Radford, Metz, and Chintala 2015)
No prior information of GANs is required – we’ll hold this submit sensible (no maths) and give attention to tips on how to obtain your objective, mapping a easy and vivid idea into an astonishingly small variety of traces of code.

As within the submit on machine translation with consideration, we first should cowl some conditions.
By the way in which, no want to repeat out the code snippets – you’ll discover the entire code in eager_dcgan.R).

Stipulations

The code on this submit relies on the most recent CRAN variations of a number of of the TensorFlow R packages. You’ll be able to set up these packages as follows:

set up.packages(c("tensorflow", "keras", "tfdatasets"))

library(tensorflow)
install_tensorflow()

We’ll additionally use the tfdatasets bundle for our enter pipeline. So we find yourself with the next preamble to set issues up:

That’s it. Let’s get began.

So what’s a GAN?

GAN stands for Generative Adversarial Community(Goodfellow et al. 2014). It’s a setup of two brokers, the generator and the discriminator, that act towards one another (thus, adversarial). It’s generative as a result of the objective is to generate output (versus, say, classification or regression).

In human studying, suggestions – direct or oblique – performs a central function. Say we wished to forge a banknote (so long as these nonetheless exist). Assuming we will get away with unsuccessful trials, we might get higher and higher at forgery over time. Optimizing our approach, we might find yourself wealthy.
This idea of optimizing from suggestions is embodied within the first of the 2 brokers, the generator. It will get its suggestions from the discriminator, in an upside-down approach: If it could idiot the discriminator, making it consider that the banknote was actual, all is okay; if the discriminator notices the pretend, it has to do issues in a different way. For a neural community, meaning it has to replace its weights.

How does the discriminator know what’s actual and what’s pretend? It too needs to be skilled, on actual banknotes (or regardless of the sort of objects concerned) and the pretend ones produced by the generator. So the entire setup is 2 brokers competing, one striving to generate realistic-looking pretend objects, and the opposite, to disavow the deception. The aim of coaching is to have each evolve and get higher, in flip inflicting the opposite to get higher, too.

On this system, there isn’t a goal minimal to the loss perform: We wish each parts to study and getter higher “in lockstep,” as a substitute of 1 successful out over the opposite. This makes optimization troublesome.
In apply due to this fact, tuning a GAN can appear extra like alchemy than like science, and it typically is sensible to lean on practices and “methods” reported by others.

On this instance, identical to within the Google pocket book we’re porting, the objective is to generate MNIST digits. Whereas that will not sound like probably the most thrilling job one might think about, it lets us give attention to the mechanics, and permits us to maintain computation and reminiscence necessities (comparatively) low.

Let’s load the information (coaching set wanted solely) after which, have a look at the primary actor in our drama, the generator.

Coaching knowledge

mnist <- dataset_mnist()
c(train_images, train_labels) %<-% mnist$practice

train_images <- train_images %>% 
  k_expand_dims() %>%
  k_cast(dtype = "float32")

# normalize photos to [-1, 1] as a result of the generator makes use of tanh activation
train_images <- (train_images - 127.5) / 127.5

Our full coaching set might be streamed as soon as per epoch:

buffer_size <- 60000
batch_size <- 256
batches_per_epoch <- (buffer_size / batch_size) %>% spherical()

train_dataset <- tensor_slices_dataset(train_images) %>%
  dataset_shuffle(buffer_size) %>%
  dataset_batch(batch_size)

This enter might be fed to the discriminator solely.

Generator

Each generator and discriminator are Keras customized fashions.
In distinction to customized layers, customized fashions let you assemble fashions as unbiased items, full with customized ahead go logic, backprop and optimization. The model-generating perform defines the layers the mannequin (self) needs assigned, and returns the perform that implements the ahead go.

As we’ll quickly see, the generator will get handed vectors of random noise for enter. This vector is reworked to 3d (peak, width, channels) after which, successively upsampled to the required output measurement of (28,28,3).

generator <-
  perform(identify = NULL) {
    keras_model_custom(identify = identify, perform(self) {
      
      self$fc1 <- layer_dense(items = 7 * 7 * 64, use_bias = FALSE)
      self$batchnorm1 <- layer_batch_normalization()
      self$leaky_relu1 <- layer_activation_leaky_relu()
      self$conv1 <-
        layer_conv_2d_transpose(
          filters = 64,
          kernel_size = c(5, 5),
          strides = c(1, 1),
          padding = "identical",
          use_bias = FALSE
        )
      self$batchnorm2 <- layer_batch_normalization()
      self$leaky_relu2 <- layer_activation_leaky_relu()
      self$conv2 <-
        layer_conv_2d_transpose(
          filters = 32,
          kernel_size = c(5, 5),
          strides = c(2, 2),
          padding = "identical",
          use_bias = FALSE
        )
      self$batchnorm3 <- layer_batch_normalization()
      self$leaky_relu3 <- layer_activation_leaky_relu()
      self$conv3 <-
        layer_conv_2d_transpose(
          filters = 1,
          kernel_size = c(5, 5),
          strides = c(2, 2),
          padding = "identical",
          use_bias = FALSE,
          activation = "tanh"
        )
      
      perform(inputs, masks = NULL, coaching = TRUE) {
        self$fc1(inputs) %>%
          self$batchnorm1(coaching = coaching) %>%
          self$leaky_relu1() %>%
          k_reshape(form = c(-1, 7, 7, 64)) %>%
          self$conv1() %>%
          self$batchnorm2(coaching = coaching) %>%
          self$leaky_relu2() %>%
          self$conv2() %>%
          self$batchnorm3(coaching = coaching) %>%
          self$leaky_relu3() %>%
          self$conv3()
      }
    })
  }

Discriminator

The discriminator is only a fairly regular convolutional community outputting a rating. Right here, utilization of “rating” as a substitute of “chance” is on function: For those who have a look at the final layer, it’s totally related, of measurement 1 however missing the same old sigmoid activation. It is because in contrast to Keras’ loss_binary_crossentropy, the loss perform we’ll be utilizing right here – tf$losses$sigmoid_cross_entropy – works with the uncooked logits, not the outputs of the sigmoid.

discriminator <-
  perform(identify = NULL) {
    keras_model_custom(identify = identify, perform(self) {
      
      self$conv1 <- layer_conv_2d(
        filters = 64,
        kernel_size = c(5, 5),
        strides = c(2, 2),
        padding = "identical"
      )
      self$leaky_relu1 <- layer_activation_leaky_relu()
      self$dropout <- layer_dropout(price = 0.3)
      self$conv2 <-
        layer_conv_2d(
          filters = 128,
          kernel_size = c(5, 5),
          strides = c(2, 2),
          padding = "identical"
        )
      self$leaky_relu2 <- layer_activation_leaky_relu()
      self$flatten <- layer_flatten()
      self$fc1 <- layer_dense(items = 1)
      
      perform(inputs, masks = NULL, coaching = TRUE) {
        inputs %>% self$conv1() %>%
          self$leaky_relu1() %>%
          self$dropout(coaching = coaching) %>%
          self$conv2() %>%
          self$leaky_relu2() %>%
          self$flatten() %>%
          self$fc1()
      }
    })
  }

Setting the scene

Earlier than we will begin coaching, we have to create the same old parts of a deep studying setup: the mannequin (or fashions, on this case), the loss perform(s), and the optimizer(s).

Mannequin creation is only a perform name, with somewhat further on prime:

generator <- generator()
discriminator <- discriminator()

# https://www.tensorflow.org/api_docs/python/tf/contrib/keen/defun
generator$name = tf$contrib$keen$defun(generator$name)
discriminator$name = tf$contrib$keen$defun(discriminator$name)

defun compiles an R perform (as soon as per completely different mixture of argument shapes and non-tensor objects values)) right into a TensorFlow graph, and is used to hurry up computations. This comes with unwanted effects and presumably surprising habits – please seek the advice of the documentation for the small print. Right here, we have been primarily curious in how a lot of a speedup we’d discover when utilizing this from R – in our instance, it resulted in a speedup of 130%.

On to the losses. Discriminator loss consists of two components: Does it accurately establish actual photos as actual, and does it accurately spot pretend photos as pretend.
Right here real_output and generated_output comprise the logits returned from the discriminator – that’s, its judgment of whether or not the respective photos are pretend or actual.

discriminator_loss <- perform(real_output, generated_output) {
  real_loss <- tf$losses$sigmoid_cross_entropy(
    multi_class_labels = k_ones_like(real_output),
    logits = real_output)
  generated_loss <- tf$losses$sigmoid_cross_entropy(
    multi_class_labels = k_zeros_like(generated_output),
    logits = generated_output)
  real_loss + generated_loss
}

Generator loss relies on how the discriminator judged its creations: It might hope for all of them to be seen as actual.

generator_loss <- perform(generated_output) {
  tf$losses$sigmoid_cross_entropy(
    tf$ones_like(generated_output),
    generated_output)
}

Now we nonetheless have to outline optimizers, one for every mannequin.

discriminator_optimizer <- tf$practice$AdamOptimizer(1e-4)
generator_optimizer <- tf$practice$AdamOptimizer(1e-4)

Coaching loop

There are two fashions, two loss capabilities and two optimizers, however there is only one coaching loop, as each fashions depend upon one another.
The coaching loop might be over MNIST photos streamed in batches, however we nonetheless want enter to the generator – a random vector of measurement 100, on this case.

Let’s take the coaching loop step-by-step.
There might be an outer and an internal loop, one over epochs and one over batches.
At the beginning of every epoch, we create a recent iterator over the dataset:

for (epoch in seq_len(num_epochs)) {
  begin <- Sys.time()
  total_loss_gen <- 0
  total_loss_disc <- 0
  iter <- make_iterator_one_shot(train_dataset)

until_out_of_range({
  batch <- iterator_get_next(iter)
  noise <- k_random_normal(c(batch_size, noise_dim))
  with(tf$GradientTape() %as% gen_tape, { with(tf$GradientTape() %as% disc_tape, {
    generated_images <- generator(noise)
    disc_real_output <- discriminator(batch, coaching = TRUE)
    disc_generated_output <-
       discriminator(generated_images, coaching = TRUE)
    gen_loss <- generator_loss(disc_generated_output)
    disc_loss <- discriminator_loss(disc_real_output, disc_generated_output)
  }) })

gradients_of_generator <-
  gen_tape$gradient(gen_loss, generator$variables)
gradients_of_discriminator <-
  disc_tape$gradient(disc_loss, discriminator$variables)
      
generator_optimizer$apply_gradients(purrr::transpose(
  record(gradients_of_generator, generator$variables)
))
discriminator_optimizer$apply_gradients(purrr::transpose(
  record(gradients_of_discriminator, discriminator$variables)
))
      
total_loss_gen <- total_loss_gen + gen_loss
total_loss_disc <- total_loss_disc + disc_loss

This ends the loop over batches. End off the loop over epochs displaying present losses and saving a number of of the generator’s paintings:

cat("Time for epoch ", epoch, ": ", Sys.time() - begin, "n")
cat("Generator loss: ", total_loss_gen$numpy() / batches_per_epoch, "n")
cat("Discriminator loss: ", total_loss_disc$numpy() / batches_per_epoch, "nn")
if (epoch %% 10 == 0)
  generate_and_save_images(generator,
                           epoch,
                           random_vector_for_generation)

Right here’s the coaching loop once more, proven as a complete – even together with the traces for reporting on progress, it’s remarkably concise, and permits for a fast grasp of what’s going on:

practice <- perform(dataset, epochs, noise_dim) {
  for (epoch in seq_len(num_epochs)) {
    begin <- Sys.time()
    total_loss_gen <- 0
    total_loss_disc <- 0
    iter <- make_iterator_one_shot(train_dataset)
    
    until_out_of_range({
      batch <- iterator_get_next(iter)
      noise <- k_random_normal(c(batch_size, noise_dim))
      with(tf$GradientTape() %as% gen_tape, { with(tf$GradientTape() %as% disc_tape, {
        generated_images <- generator(noise)
        disc_real_output <- discriminator(batch, coaching = TRUE)
        disc_generated_output <-
          discriminator(generated_images, coaching = TRUE)
        gen_loss <- generator_loss(disc_generated_output)
        disc_loss <-
          discriminator_loss(disc_real_output, disc_generated_output)
      }) })
      
      gradients_of_generator <-
        gen_tape$gradient(gen_loss, generator$variables)
      gradients_of_discriminator <-
        disc_tape$gradient(disc_loss, discriminator$variables)
      
      generator_optimizer$apply_gradients(purrr::transpose(
        record(gradients_of_generator, generator$variables)
      ))
      discriminator_optimizer$apply_gradients(purrr::transpose(
        record(gradients_of_discriminator, discriminator$variables)
      ))
      
      total_loss_gen <- total_loss_gen + gen_loss
      total_loss_disc <- total_loss_disc + disc_loss
      
    })
    
    cat("Time for epoch ", epoch, ": ", Sys.time() - begin, "n")
    cat("Generator loss: ", total_loss_gen$numpy() / batches_per_epoch, "n")
    cat("Discriminator loss: ", total_loss_disc$numpy() / batches_per_epoch, "nn")
    if (epoch %% 10 == 0)
      generate_and_save_images(generator,
                               epoch,
                               random_vector_for_generation)
    
  }
}

Right here’s the perform for saving generated photos…

generate_and_save_images <- perform(mannequin, epoch, test_input) {
  predictions <- mannequin(test_input, coaching = FALSE)
  png(paste0("images_epoch_", epoch, ".png"))
  par(mfcol = c(5, 5))
  par(mar = c(0.5, 0.5, 0.5, 0.5),
      xaxs = 'i',
      yaxs = 'i')
  for (i in 1:25) {
    img <- predictions[i, , , 1]
    img <- t(apply(img, 2, rev))
    picture(
      1:28,
      1:28,
      img * 127.5 + 127.5,
      col = grey((0:255) / 255),
      xaxt = 'n',
      yaxt = 'n'
    )
  }
  dev.off()
}

… and we’re able to go!

num_epochs <- 150
practice(train_dataset, num_epochs, noise_dim)

Outcomes

Listed below are some generated photos after coaching for 150 epochs:

As they are saying, your outcomes will most actually fluctuate!

Conclusion

Whereas actually tuning GANs will stay a problem, we hope we have been capable of present that mapping ideas to code just isn’t troublesome when utilizing keen execution. In case you’ve performed round with GANs earlier than, you’ll have discovered you wanted to pay cautious consideration to arrange the losses the suitable approach, freeze the discriminator’s weights when wanted, and so on. This want goes away with keen execution.
In upcoming posts, we’ll present additional examples the place utilizing it makes mannequin improvement simpler.