cleanup

neural_stylization/loss_functions.py

@@ -2,6 +2,22 @@
 from keras import backend as K
 
 
+def content_loss(content, combination):
+    """
+    Return the content loss between the content and combinations tensors.
+
+    Args:
+        content: the output of a layer for the content image
+        combination: the output of a layer for the combination image
+
+    Returns:
+        the loss between `content` and `combination`
+
+    """
+    # squared euclidean distance, exactly how it is in the paper
+    return 0.5 * K.sum(K.square(combination - content))
+
+
 def gram(matrix):
     """
     Return a gram matrix for the given input matrix.
@@ -24,22 +40,6 @@ def gram(matrix):
     return g
 
 
-def content_loss(content, combination):
-    """
-    Return the content loss between the content and combinations tensors.
-
-    Args:
-        content: the output of a layer for the content image
-        combination: the output of a layer for the combination image
-
-    Returns:
-        the loss between `content` and `combination`
-
-    """
-    # squared euclidean distance, exactly how it is in the paper
-    return 0.5 * K.sum(K.square(combination - content))
-
-
 def style_loss(style, combination):
     """
     Return the style loss between the style and combinations tensors.
@@ -63,45 +63,4 @@ def style_loss(style, combination):
     return K.sum(K.square(gram(style) - gram(combination))) / (4 * Nl_2 * Ml_2)
 
 
-def total_variation_loss(combination, kind='anisotropic'):
-    """
-    Return the total variation loss for the combination image.
-
-    Args:
-        combination: the combination tensor to return the variation loss of
-        kind: the kind of total variation loss to use (default 'anisotropic')
-
-    Returns:
-        the total variation loss of the combination tensor
-
-    """
-    # store the dimensions for indexing from thee combination
-    h, w = combination.shape[1], combination.shape[2]
-    if kind == 'anisotropic':
-        # take the absolute value between this image, and the image one pixel
-        # down, and one pixel to the right. take the absolute value as
-        # specified by anisotropic loss
-        a = K.abs(combination[:, :h-1, :w-1, :] - combination[:, 1:, :w-1, :])
-        b = K.abs(combination[:, :h-1, :w-1, :] - combination[:, :h-1, 1:, :])
-        # add up all the differences
-        return K.sum(a + b)
-    elif kind == 'isotropic':
-        # take the absolute value between this image, and the image one pixel
-        # down, and one pixel to the right. take the square root as specified
-        # by isotropic loss
-        a = K.square(combination[:, :h-1, :w-1, :] - combination[:, 1:, :w-1, :])
-        b = K.square(combination[:, :h-1, :w-1, :] - combination[:, :h-1, 1:, :])
-        # take the vector square root of all the pixel differences, then sum
-        # them all up
-        return K.sum(K.pow(a + b, 2))
-    else:
-        # kind can only be two values, raise an error on unexpected kind value
-        raise ValueError("`kind` should be 'anisotropic' or 'isotropic'")
-
-
-# explicitly export the public API
-__all__ = [
-    'content_loss',
-    'style_loss',
-    'total_variation_loss'
-]
+__all__ = ['content_loss', 'style_loss']

neural_stylization/optimizers/__init__.py

@@ -1,3 +1,3 @@
-"""Black Box optimizers for the project."""
+"""Black Box optimization methods."""
 from .sgd import SGD
 from .l_bfgs import L_BFGS

neural_stylization/optimizers/l_bfgs.py

@@ -1,6 +1,7 @@
 """An interface to the L-BFGS Algorithm."""
 import numpy as np
 from typing import Callable
+from tqdm import tqdm
 from scipy.optimize import fmin_l_bfgs_b
 
 
@@ -75,7 +76,7 @@ def loss_and_gradients(X):
         # assign the custom method to self for loss / gradient calculation
         self.loss_and_gradients = loss_and_gradients
 
-        for i in range(iterations):
+        for i in tqdm(range(iterations)):
             # pass X through an iteration of LBFGS
             X, min_val, info = fmin_l_bfgs_b(self.loss, X,
                                              fprime=self.gradients,

neural_stylization/optimizers/sgd.py

@@ -1,5 +1,6 @@
 """An implementation of a basic gradient descent algorithm."""
 from typing import Callable
+from tqdm import tqdm
 import numpy as np
 
 
@@ -43,7 +44,7 @@ def minimize(self,
 
         Returns: an optimized X about the loss and gradients given
         """
-        for i in range(iterations):
+        for i in tqdm(range(iterations)):
             # pass the input through the loss function and generate gradients
             loss_i, grads_i = loss_grads([X])
             # move the input based on the gradients and learning rate

neural_stylization/reconstruct_content.py

@@ -4,11 +4,11 @@
 from typing import Callable
 from keras import backend as K
 from .vgg19 import VGG_19
-from .img_util import load_image
-from .img_util import normalize
-from .img_util import denormalize
-from .img_util import image_to_matrix
-from .img_util import matrix_to_image
+from .util.img_util import load_image
+from .util.img_util import normalize
+from .util.img_util import denormalize
+from .util.img_util import image_to_matrix
+from .util.img_util import matrix_to_image
 from .loss_functions import content_loss
 from .optimizers.l_bfgs import L_BFGS
 

neural_stylization/reconstruct_style.py

@@ -4,11 +4,11 @@
 from typing import Callable
 from keras import backend as K
 from .vgg19 import VGG_19
-from .img_util import load_image
-from .img_util import normalize
-from .img_util import denormalize
-from .img_util import image_to_matrix
-from .img_util import matrix_to_image
+from .util.img_util import load_image
+from .util.img_util import normalize
+from .util.img_util import denormalize
+from .util.img_util import image_to_matrix
+from .util.img_util import matrix_to_image
 from .loss_functions import style_loss
 from .optimizers.l_bfgs import L_BFGS
 

neural_stylization/transfer_style.py

@@ -3,13 +3,13 @@
 from PIL import Image
 from typing import Callable
 from keras import backend as K
-from neural_stylization.vgg19 import VGG_19
-from neural_stylization.img_util import normalize
-from neural_stylization.img_util import denormalize
-from neural_stylization.img_util import load_image
-from neural_stylization.img_util import image_to_matrix
-from neural_stylization.img_util import matrix_to_image
-from neural_stylization.loss_functions import content_loss, style_loss
+from .vgg19 import VGG_19
+from .util.img_util import normalize
+from .util.img_util import denormalize
+from .util.img_util import load_image
+from .util.img_util import image_to_matrix
+from .util.img_util import matrix_to_image
+from .loss_functions import content_loss, style_loss
 
 
 # the template for the class's repr method

neural_stylization/util/__init__.py

@@ -0,0 +1 @@
+"""Utility methods for the parent package."""

neural_stylization/build_callback.py → neural_stylization/util/build_callback.py

@@ -20,8 +20,9 @@ def build_callback(out_dir: str):
     for file in glob('{}/*.png'.format(out_dir)):
         os.remove(file)
 
-    from neural_stylization.img_util import denormalize
-    from neural_stylization.img_util import matrix_to_image
+    from .img_util import denormalize
+    from .img_util import matrix_to_image
+    from IPython import display
 
     def denormalize_and_display(image, i) -> None:
         """
@@ -33,7 +34,6 @@ def denormalize_and_display(image, i) -> None:
 
         Returns: None
         """
-        from IPython import display
         # clear the existing output
         display.clear_output(wait=True)
         # denormalize the image and conver to binary
@@ -42,7 +42,5 @@ def denormalize_and_display(image, i) -> None:
         image.save('{}/{}.png'.format(out_dir, i))
         # display the image on the IPython front end
         display.display(image)
-        # display the iteration beneat the image
-        display.display(i)
 
     return denormalize_and_display

neural_stylization/vgg19.py

@@ -16,9 +16,7 @@
 from keras import backend as K
 
 
-# the definition for a tensor type as defined by the keras documentation:
-# https://keras.io/backend/
-# scroll to the `is_keras_tensor` section for a description of this type
+# the definition for a tensor type as defined by Keras
 Tensor = Union[Input, Layer]