Custom resizing function in flow from directory

keras_preprocessing/image/dataframe_iterator.py

@@ -80,6 +80,11 @@ class DataFrameIterator(BatchFromFilesMixin, Iterator):
             If PIL version 1.1.3 or newer is installed, "lanczos" is also
             supported. If PIL version 3.4.0 or newer is installed, "box" and
             "hamming" are also supported. By default, "nearest" is used.
+        resizing_function: function, used to resize the loaded images to the
+            target size. this will overrule interpolation. If None, then
+            interpolation will happen. The input is an image in the specified
+            data format, and the output has to be an image in the specified
+            data format with the target size.
         dtype: Dtype to use for the generated arrays.
         validate_filenames: Boolean, whether to validate image filenames in
         `x_col`. If `True`, invalid images will be ignored. Disabling this option
@@ -109,6 +114,7 @@ def __init__(self,
                  save_format='png',
                  subset=None,
                  interpolation='nearest',
+                 resizing_function=None,
                  dtype='float32',
                  validate_filenames=True):
 
@@ -120,7 +126,8 @@ def __init__(self,
                                                             save_prefix,
                                                             save_format,
                                                             subset,
-                                                            interpolation)
+                                                            interpolation,
+                                                            resizing_function)
         df = dataframe.copy()
         self.directory = directory or ''
         self.class_mode = class_mode

keras_preprocessing/image/directory_iterator.py

@@ -60,6 +60,11 @@ class DirectoryIterator(BatchFromFilesMixin, Iterator):
             If PIL version 1.1.3 or newer is installed, "lanczos" is also
             supported. If PIL version 3.4.0 or newer is installed, "box" and
             "hamming" are also supported. By default, "nearest" is used.
+        resizing_function: function, used to resize the loaded images to the
+            target size. this will overrule interpolation. If None, then
+            interpolation will happen. The input is an image in the specified
+            data format, and the output has to be an image in the specified
+            data format with the target size.
         dtype: Dtype to use for generated arrays.
     """
     allowed_class_modes = {'categorical', 'binary', 'sparse', 'input', None}
@@ -81,6 +86,7 @@ def __init__(self,
                  follow_links=False,
                  subset=None,
                  interpolation='nearest',
+                 resizing_function=None,
                  dtype='float32'):
         super(DirectoryIterator, self).set_processing_attrs(image_data_generator,
                                                             target_size,
@@ -90,7 +96,8 @@ def __init__(self,
                                                             save_prefix,
                                                             save_format,
                                                             subset,
-                                                            interpolation)
+                                                            interpolation,
+                                                            resizing_function)
         self.directory = directory
         self.classes = classes
         if class_mode not in self.allowed_class_modes:

keras_preprocessing/image/image_data_generator.py

@@ -447,7 +447,8 @@ def flow_from_directory(self,
                             save_format='png',
                             follow_links=False,
                             subset=None,
-                            interpolation='nearest'):
+                            interpolation='nearest',
+                            resizing_function=None):
         """Takes the path to a directory & generates batches of augmented data.
 
         # Arguments
@@ -515,6 +516,11 @@ class subdirectories (default: False).
                 supported. If PIL version 3.4.0 or newer is installed,
                 `"box"` and `"hamming"` are also supported.
                 By default, `"nearest"` is used.
+            resizing_function: function, used to resize the loaded images to the
+                target size. this will overrule interpolation. If None, then
+                interpolation will happen. The input is an image in the specified
+                data format, and the output has to be an image in the specified
+                data format with the target size.
 
         # Returns
             A `DirectoryIterator` yielding tuples of `(x, y)`
@@ -539,7 +545,8 @@ class subdirectories (default: False).
             follow_links=follow_links,
             subset=subset,
             interpolation=interpolation,
-            dtype=self.dtype
+            dtype=self.dtype,
+            resizing_function=resizing_function
         )
 
     def flow_from_dataframe(self,

keras_preprocessing/image/iterator.py

@@ -142,7 +142,8 @@ def set_processing_attrs(self,
                              save_prefix,
                              save_format,
                              subset,
-                             interpolation):
+                             interpolation,
+                             resizing_function):
         """Sets attributes to use later for processing files into a batch.
 
         # Arguments
@@ -168,6 +169,11 @@ def set_processing_attrs(self,
                 If PIL version 1.1.3 or newer is installed, "lanczos" is also
                 supported. If PIL version 3.4.0 or newer is installed, "box" and
                 "hamming" are also supported. By default, "nearest" is used.
+            resizing_function: function, used to resize the loaded images to the
+                target size. this will overrule interpolation. If None, then
+                interpolation will happen. The input is an image in the specified
+                data format, and the output has to be an image in the specified
+                data format with the target size.
         """
         self.image_data_generator = image_data_generator
         self.target_size = tuple(target_size)
@@ -195,6 +201,7 @@ def set_processing_attrs(self,
         self.save_prefix = save_prefix
         self.save_format = save_format
         self.interpolation = interpolation
+        self.resizing_function = resizing_function
         if subset is not None:
             validation_split = self.image_data_generator._validation_split
             if subset == 'validation':
@@ -223,12 +230,24 @@ def _get_batches_of_transformed_samples(self, index_array):
         # build batch of image data
         # self.filepaths is dynamic, is better to call it once outside the loop
         filepaths = self.filepaths
+        load_target_size = self.target_size
+        if self.resizing_function is not None:
+            load_target_size = None
         for i, j in enumerate(index_array):
             img = load_img(filepaths[j],
                            color_mode=self.color_mode,
-                           target_size=self.target_size,
+                           target_size=load_target_size,
                            interpolation=self.interpolation)
             x = img_to_array(img, data_format=self.data_format)
+            # NOTE: we could potentially have keyword arguments for the
+            # resizing function
+            if self.resizing_function is not None:
+                x = self.resizing_function(x)
+            if x.shape != self.image_shape:
+                raise ValueError(
+                    'The loaded image shape %s (at %s) does not correspond to'
+                    'the specified image shape %s' %
+                    (str(x.shape), filepaths[j], str(self.image_shape)))
             # Pillow images should be closed after `load_img`,
             # but not PIL images.
             if hasattr(img, 'close'):