Neural type switch with keen execution and Keras

How would your summer season vacation’s images look had Edvard Munch painted them? (Maybe it’s higher to not know).
Let’s take a extra comforting instance: How would a pleasant, summarly river panorama look if painted by Katsushika Hokusai?

Type switch on photos just isn’t new, however obtained a lift when Gatys, Ecker, and Bethge(Gatys, Ecker, and Bethge 2015) confirmed easy methods to efficiently do it with deep studying.
The primary concept is simple: Create a hybrid that may be a tradeoff between the content material picture we wish to manipulate, and a type picture we wish to imitate, by optimizing for maximal resemblance to each on the identical time.

For those who’ve learn the chapter on neural type switch from Deep Studying with R, chances are you’ll acknowledge a few of the code snippets that comply with.
Nonetheless, there is a crucial distinction: This publish makes use of TensorFlow Keen Execution, permitting for an crucial manner of coding that makes it straightforward to map ideas to code.
Identical to earlier posts on keen execution on this weblog, this can be a port of a Google Colaboratory pocket book that performs the identical activity in Python.

As regular, please be sure to have the required bundle variations put in. And no want to repeat the snippets – you’ll discover the entire code among the many Keras examples.

Stipulations

The code on this publish relies on the latest variations of a number of of the TensorFlow R packages. You possibly can set up these packages as follows:

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

library(tensorflow)
install_tensorflow()

There are extra necessities for utilizing TensorFlow keen execution. First, we have to name tfe_enable_eager_execution() proper originally of this system. Second, we have to use the implementation of Keras included in TensorFlow, somewhat than the bottom Keras implementation.

Stipulations behind us, let’s get began!

Enter photos

Right here is our content material picture – change by a picture of your personal:

# In case you have sufficient reminiscence in your GPU, no must load the photographs
# at such small measurement.
# That is the dimensions I discovered working for a 4G GPU.
img_shape <- c(128, 128, 3)

content_path <- "isar.jpg"

content_image <-  image_load(content_path, target_size = img_shape[1:2])
content_image %>% 
  image_to_array() %>%
  `/`(., 255) %>%
  as.raster() %>%
  plot()

And right here’s the type mannequin, Hokusai’s The Nice Wave off Kanagawa, which you’ll obtain from Wikimedia Commons:

style_path <- "The_Great_Wave_off_Kanagawa.jpg"

style_image <-  image_load(content_path, target_size = img_shape[1:2])
style_image %>% 
  image_to_array() %>%
  `/`(., 255) %>%
  as.raster() %>%
  plot()

We create a wrapper that masses and preprocesses the enter photos for us.
As we might be working with VGG19, a community that has been educated on ImageNet, we have to remodel our enter photos in the identical manner that was used coaching it. Later, we’ll apply the inverse transformation to our mixture picture earlier than displaying it.

load_and_preprocess_image <- perform(path) {
  img <- image_load(path, target_size = img_shape[1:2]) %>%
    image_to_array() %>%
    k_expand_dims(axis = 1) %>%
    imagenet_preprocess_input()
}

deprocess_image <- perform(x) {
  x <- x[1, , ,]
  # Take away zero-center by imply pixel
  x[, , 1] <- x[, , 1] + 103.939
  x[, , 2] <- x[, , 2] + 116.779
  x[, , 3] <- x[, , 3] + 123.68
  # 'BGR'->'RGB'
  x <- x[, , c(3, 2, 1)]
  x[x > 255] <- 255
  x[x < 0] <- 0
  x[] <- as.integer(x) / 255
  x
}

Setting the scene

We’re going to use a neural community, however we gained’t be coaching it. Neural type switch is a bit unusual in that we don’t optimize the community’s weights, however again propagate the loss to the enter layer (the picture), with a purpose to transfer it within the desired course.

We might be excited about two sorts of outputs from the community, comparable to our two objectives.
Firstly, we wish to preserve the mixture picture just like the content material picture, on a excessive degree. In a convnet, higher layers map to extra holistic ideas, so we’re choosing a layer excessive up within the graph to check outputs from the supply and the mixture.

Secondly, the generated picture ought to “appear like” the type picture. Type corresponds to decrease degree options like texture, shapes, strokes… So to check the mixture in opposition to the type instance, we select a set of decrease degree conv blocks for comparability and combination the outcomes.

content_layers <- c("block5_conv2")
style_layers <- c("block1_conv1",
                 "block2_conv1",
                 "block3_conv1",
                 "block4_conv1",
                 "block5_conv1")

num_content_layers <- size(content_layers)
num_style_layers <- size(style_layers)

get_model <- perform() {
  vgg <- application_vgg19(include_top = FALSE, weights = "imagenet")
  vgg$trainable <- FALSE
  style_outputs <- map(style_layers, perform(layer) vgg$get_layer(layer)$output)
  content_outputs <- map(content_layers, perform(layer) vgg$get_layer(layer)$output)
  model_outputs <- c(style_outputs, content_outputs)
  keras_model(vgg$enter, model_outputs)
}

Losses

When optimizing the enter picture, we’ll think about three kinds of losses. Firstly, the content material loss: How totally different is the mixture picture from the supply? Right here, we’re utilizing the sum of the squared errors for comparability.

content_loss <- perform(content_image, goal) {
  k_sum(k_square(goal - content_image))
}

Our second concern is having the types match as intently as potential. Type is often operationalized because the Gram matrix of flattened characteristic maps in a layer. We thus assume that type is expounded to how maps in a layer correlate with different.

We due to this fact compute the Gram matrices of the layers we’re excited about (outlined above), for the supply picture in addition to the optimization candidate, and evaluate them, once more utilizing the sum of squared errors.

gram_matrix <- perform(x) {
  options <- k_batch_flatten(k_permute_dimensions(x, c(3, 1, 2)))
  gram <- k_dot(options, k_transpose(options))
  gram
}

style_loss <- perform(gram_target, mixture) {
  gram_comb <- gram_matrix(mixture)
  k_sum(k_square(gram_target - gram_comb)) /
    (4 * (img_shape[3] ^ 2) * (img_shape[1] * img_shape[2]) ^ 2)
}

Thirdly, we don’t need the mixture picture to look overly pixelated, thus we’re including in a regularization part, the whole variation within the picture:

total_variation_loss <- perform(picture) {
  y_ij  <- picture[1:(img_shape[1] - 1L), 1:(img_shape[2] - 1L),]
  y_i1j <- picture[2:(img_shape[1]), 1:(img_shape[2] - 1L),]
  y_ij1 <- picture[1:(img_shape[1] - 1L), 2:(img_shape[2]),]
  a <- k_square(y_ij - y_i1j)
  b <- k_square(y_ij - y_ij1)
  k_sum(k_pow(a + b, 1.25))
}

The difficult factor is easy methods to mix these losses. We’ve reached acceptable outcomes with the next weightings, however be at liberty to mess around as you see match:

content_weight <- 100
style_weight <- 0.8
total_variation_weight <- 0.01

Get mannequin outputs for the content material and magnificence photos

We want the mannequin’s output for the content material and magnificence photos, however right here it suffices to do that simply as soon as.
We concatenate each photos alongside the batch dimension, cross that enter to the mannequin, and get again an inventory of outputs, the place each ingredient of the listing is a 4-d tensor. For the type picture, we’re within the type outputs at batch place 1, whereas for the content material picture, we want the content material output at batch place 2.

Within the under feedback, please word that the sizes of dimensions 2 and three will differ should you’re loading photos at a distinct measurement.

get_feature_representations <-
  perform(mannequin, content_path, style_path) {
    
    # dim == (1, 128, 128, 3)
    style_image <-
      load_and_process_image(style_path) %>% k_cast("float32")
    # dim == (1, 128, 128, 3)
    content_image <-
      load_and_process_image(content_path) %>% k_cast("float32")
    # dim == (2, 128, 128, 3)
    stack_images <- k_concatenate(listing(style_image, content_image), axis = 1)
    
    # size(model_outputs) == 6
    # dim(model_outputs[[1]]) = (2, 128, 128, 64)
    # dim(model_outputs[[6]]) = (2, 8, 8, 512)
    model_outputs <- mannequin(stack_images)
    
    style_features <- 
      model_outputs[1:num_style_layers] %>%
      map(perform(batch) batch[1, , , ])
    content_features <- 
      model_outputs[(num_style_layers + 1):(num_style_layers + num_content_layers)] %>%
      map(perform(batch) batch[2, , , ])
    
    listing(style_features, content_features)
  }

Computing the losses

On each iteration, we have to cross the mixture picture by way of the mannequin, get hold of the type and content material outputs, and compute the losses. Once more, the code is extensively commented with tensor sizes for straightforward verification, however please take into account that the precise numbers presuppose you’re working with 128×128 photos.

compute_loss <-
  perform(mannequin, loss_weights, init_image, gram_style_features, content_features) {
    
    c(style_weight, content_weight) %<-% loss_weights
    model_outputs <- mannequin(init_image)
    style_output_features <- model_outputs[1:num_style_layers]
    content_output_features <-
      model_outputs[(num_style_layers + 1):(num_style_layers + num_content_layers)]
    
    # type loss
    weight_per_style_layer <- 1 / num_style_layers
    style_score <- 0
    # dim(style_zip[[5]][[1]]) == (512, 512)
    style_zip <- transpose(listing(gram_style_features, style_output_features))
    for (l in 1:size(style_zip)) {
      # for l == 1:
      # dim(target_style) == (64, 64)
      # dim(comb_style) == (1, 128, 128, 64)
      c(target_style, comb_style) %<-% style_zip[[l]]
      style_score <- style_score + weight_per_style_layer * 
        style_loss(target_style, comb_style[1, , , ])
    }
    
    # content material loss
    weight_per_content_layer <- 1 / num_content_layers
    content_score <- 0
    content_zip <- transpose(listing(content_features, content_output_features))
    for (l in 1:size(content_zip)) {
      # dim(comb_content) ==  (1, 8, 8, 512)
      # dim(target_content) == (8, 8, 512)
      c(target_content, comb_content) %<-% content_zip[[l]]
      content_score <- content_score + weight_per_content_layer *
        content_loss(comb_content[1, , , ], target_content)
    }
    
    # complete variation loss
    variation_loss <- total_variation_loss(init_image[1, , ,])
    
    style_score <- style_score * style_weight
    content_score <- content_score * content_weight
    variation_score <- variation_loss * total_variation_weight
    
    loss <- style_score + content_score + variation_score
    listing(loss, style_score, content_score, variation_score)
  }

Computing the gradients

As quickly as we have now the losses, acquiring the gradients of the general loss with respect to the enter picture is only a matter of calling tape$gradient on the GradientTape. Observe that the nested name to compute_loss, and thus the decision of the mannequin on our mixture picture, occurs contained in the GradientTape context.

compute_grads <- 
  perform(mannequin, loss_weights, init_image, gram_style_features, content_features) {
    with(tf$GradientTape() %as% tape, {
      scores <-
        compute_loss(mannequin,
                     loss_weights,
                     init_image,
                     gram_style_features,
                     content_features)
    })
    total_loss <- scores[[1]]
    listing(tape$gradient(total_loss, init_image), scores)
  }

Coaching part

Now it’s time to coach! Whereas the pure continuation of this sentence would have been “… the mannequin,” the mannequin we’re coaching right here just isn’t VGG19 (that one we’re simply utilizing as a device), however a minimal setup of simply:

a Variable that holds our to-be-optimized picture
the loss capabilities we outlined above
an optimizer that may apply the calculated gradients to the picture variable (tf$prepare$AdamOptimizer)

Beneath, we get the type options (of the type picture) and the content material characteristic (of the content material picture) simply as soon as, then iterate over the optimization course of, saving the output each 100 iterations.

In distinction to the unique article and the Deep Studying with R e-book, however following the Google pocket book as a substitute, we’re not utilizing L-BFGS for optimization, however Adam, as our aim right here is to offer a concise introduction to keen execution.
Nonetheless, you might plug in one other optimization methodology should you wished, changing
optimizer$apply_gradients(listing(tuple(grads, init_image)))
by an algorithm of your alternative (and naturally, assigning the results of the optimization to the Variable holding the picture).

run_style_transfer <- perform(content_path, style_path) {
  mannequin <- get_model()
  stroll(mannequin$layers, perform(layer) layer$trainable = FALSE)
  
  c(style_features, content_features) %<-% 
    get_feature_representations(mannequin, content_path, style_path)
  # dim(gram_style_features[[1]]) == (64, 64)
  gram_style_features <- map(style_features, perform(characteristic) gram_matrix(characteristic))
  
  init_image <- load_and_process_image(content_path)
  init_image <- tf$contrib$keen$Variable(init_image, dtype = "float32")
  
  optimizer <- tf$prepare$AdamOptimizer(learning_rate = 1,
                                      beta1 = 0.99,
                                      epsilon = 1e-1)
  
  c(best_loss, best_image) %<-% listing(Inf, NULL)
  loss_weights <- listing(style_weight, content_weight)
  
  start_time <- Sys.time()
  global_start <- Sys.time()
  
  norm_means <- c(103.939, 116.779, 123.68)
  min_vals <- -norm_means
  max_vals <- 255 - norm_means
  
  for (i in seq_len(num_iterations)) {
    # dim(grads) == (1, 128, 128, 3)
    c(grads, all_losses) %<-% compute_grads(mannequin,
                                            loss_weights,
                                            init_image,
                                            gram_style_features,
                                            content_features)
    c(loss, style_score, content_score, variation_score) %<-% all_losses
    optimizer$apply_gradients(listing(tuple(grads, init_image)))
    clipped <- tf$clip_by_value(init_image, min_vals, max_vals)
    init_image$assign(clipped)
    
    end_time <- Sys.time()
    
    if (k_cast_to_floatx(loss) < best_loss) {
      best_loss <- k_cast_to_floatx(loss)
      best_image <- init_image
    }
    
    if (i %% 50 == 0) {
      glue("Iteration: {i}") %>% print()
      glue(
        "Complete loss: {k_cast_to_floatx(loss)},
        type loss: {k_cast_to_floatx(style_score)},
        content material loss: {k_cast_to_floatx(content_score)},
        complete variation loss: {k_cast_to_floatx(variation_score)},
        time for 1 iteration: {(Sys.time() - start_time) %>% spherical(2)}"
      ) %>% print()
      
      if (i %% 100 == 0) {
        png(paste0("style_epoch_", i, ".png"))
        plot_image <- best_image$numpy()
        plot_image <- deprocess_image(plot_image)
        plot(as.raster(plot_image), important = glue("Iteration {i}"))
        dev.off()
      }
    }
  }
  
  glue("Complete time: {Sys.time() - global_start} seconds") %>% print()
  listing(best_image, best_loss)
}

Able to run

Now, we’re prepared to start out the method:

c(best_image, best_loss) %<-% run_style_transfer(content_path, style_path)

In our case, outcomes didn’t change a lot after ~ iteration 1000, and that is how our river panorama was wanting:

… positively extra inviting than had it been painted by Edvard Munch!

Conclusion

With neural type switch, some fiddling round could also be wanted till you get the outcome you need. However as our instance reveals, this doesn’t imply the code must be sophisticated. Moreover to being straightforward to understand, keen execution additionally allows you to add debugging output, and step by way of the code line-by-line to examine on tensor shapes.
Till subsequent time in our keen execution collection!

Gatys, Leon A., Alexander S. Ecker, and Matthias Bethge. 2015. “A Neural Algorithm of Inventive Type.” CoRR abs/1508.06576. http://arxiv.org/abs/1508.06576.