Fix tf.where inconsistent dtype bug (#126)

* Make sure `tf.where` having same dtype for x and y

* Upgrade to `0.5.8`

keras_aug/__init__.py

@@ -8,4 +8,4 @@
 from keras_aug.core import SignedNormalFactorSampler
 from keras_aug.core import UniformFactorSampler
 
-__version__ = "0.5.7"
+__version__ = "0.5.8"

keras_aug/datapoints/bounding_box/utils.py

@@ -19,7 +19,9 @@ def _relative_area(boxes, bounding_box_format):
     heights = boxes[..., 3]
     # handle corner case where shear performs a full inversion.
     return tf.where(
-        tf.math.logical_and(widths > 0, heights > 0), widths * heights, 0.0
+        tf.math.logical_and(widths > 0, heights > 0),
+        widths * heights,
+        tf.constant(0.0, dtype=widths.dtype),
     )
 
 

keras_aug/layers/augmentation/geometry/random_zoom_and_crop.py

@@ -154,22 +154,22 @@ def get_random_transformation_batch(
         tops = tf.where(
             new_heights < self.crop_height,
             tf.cast((self.crop_height - new_heights) / 2, tf.int32),
-            0,
+            tf.constant(0, dtype=tf.int32),
         )
         bottoms = tf.where(
             new_heights < self.crop_height,
             self.crop_height - new_heights - tops,
-            0,
+            tf.constant(0, dtype=tf.int32),
         )
         lefts = tf.where(
             new_widths < self.crop_width,
             tf.cast((self.crop_width - new_widths) / 2, tf.int32),
-            0,
+            tf.constant(0, dtype=tf.int32),
         )
         rights = tf.where(
             new_widths < self.crop_width,
             self.crop_width - new_widths - lefts,
-            0,
+            tf.constant(0, dtype=tf.int32),
         )
         (tops, bottoms, lefts, rights) = augmentation_utils.get_position_params(
             tops, bottoms, lefts, rights, self.position, self._random_generator

keras_aug/layers/augmentation/regularization/random_cutout.py

@@ -195,7 +195,7 @@ def augment_bounding_boxes(
         # bounding_box.to_ragged() after self.augment_bounding_boxes()
         bounding_boxes["classes"] = tf.where(
             intersection_ratios >= self.bbox_removal_threshold,
-            -1.0,
+            tf.constant(-1.0, dtype=bounding_boxes["classes"].dtype),
             bounding_boxes["classes"],
         )
         return bounding_boxes

keras_aug/layers/preprocessing/geometry/resize.py

@@ -111,36 +111,44 @@ def get_random_transformation_batch(
             tops = tf.where(
                 new_heights > self.height,
                 tf.cast((new_heights - self.height) / 2, tf.int32),
-                0,
+                tf.constant(0, tf.int32),
             )
             bottoms = tf.where(
-                new_heights > self.height, new_heights - self.height - tops, 0
+                new_heights > self.height,
+                new_heights - self.height - tops,
+                tf.constant(0, tf.int32),
             )
             lefts = tf.where(
                 new_widths > self.width,
                 tf.cast((new_widths - self.width) / 2, tf.int32),
-                0,
+                tf.constant(0, tf.int32),
             )
             rights = tf.where(
-                new_widths > self.width, new_widths - self.width - lefts, 0
+                new_widths > self.width,
+                new_widths - self.width - lefts,
+                tf.constant(0, tf.int32),
             )
         else:
             assert self.pad_to_aspect_ratio
             tops = tf.where(
                 new_heights < self.height,
                 tf.cast((self.height - new_heights) / 2, tf.int32),
-                0,
+                tf.constant(0, tf.int32),
             )
             bottoms = tf.where(
-                new_heights < self.height, self.height - new_heights - tops, 0
+                new_heights < self.height,
+                self.height - new_heights - tops,
+                tf.constant(0, tf.int32),
             )
             lefts = tf.where(
                 new_widths < self.width,
                 tf.cast((self.width - new_widths) / 2, tf.int32),
-                0,
+                tf.constant(0, tf.int32),
             )
             rights = tf.where(
-                new_widths < self.width, self.width - new_widths - lefts, 0
+                new_widths < self.width,
+                self.width - new_widths - lefts,
+                tf.constant(0, tf.int32),
             )
         (tops, bottoms, lefts, rights) = augmentation_utils.get_position_params(
             tops, bottoms, lefts, rights, self.position, self._random_generator

keras_aug/layers/preprocessing/intensity/auto_contrast.py

@@ -47,8 +47,8 @@ def augment_images(self, images, transformations, **kwargs):
         highs = tf.reduce_max(images, axis=(1, 2), keepdims=True)
         scales = 255.0 / (highs - lows)
         eq_idxs = tf.math.is_inf(scales)
-        lows = tf.where(eq_idxs, 0.0, lows)
-        scales = tf.where(eq_idxs, 1.0, scales)
+        lows = tf.where(eq_idxs, tf.constant(0.0, dtype=lows.dtype), lows)
+        scales = tf.where(eq_idxs, tf.constant(1.0, dtype=scales.dtype), scales)
         images = tf.clip_by_value((images - lows) * scales, 0, 255)
         images = image_utils.transform_value_range(
             images,

keras_aug/layers/preprocessing/intensity/equalize.py

@@ -86,7 +86,7 @@ def equalize_single_channel(self, image, channel_index):
         big_number = 1410065408
         histogram_without_zeroes = tf.where(
             tf.equal(histogram, 0),
-            big_number,
+            tf.constant(big_number, dtype=histogram.dtype),
             histogram,
         )
         step = (