Add a condition to verify training status during image processing

keras/src/layers/preprocessing/image_preprocessing/mix_up.py

@@ -66,36 +66,38 @@ def get_random_transformation(self, data, training=True, seed=None):
         }
 
     def transform_images(self, images, transformation=None, training=True):
+        if training:
+            images = self._mix_up_input(images, transformation)
+        return images
+
+    def _mix_up_input(self, images, transformation):
         images = self.backend.cast(images, self.compute_dtype)
         mix_weight = transformation["mix_weight"]
         permutation_order = transformation["permutation_order"]
-
         mix_weight = self.backend.cast(
             self.backend.numpy.reshape(mix_weight, [-1, 1, 1, 1]),
             dtype=self.compute_dtype,
         )
-
         mix_up_images = self.backend.cast(
             self.backend.numpy.take(images, permutation_order, axis=0),
             dtype=self.compute_dtype,
         )
-
         images = mix_weight * images + (1.0 - mix_weight) * mix_up_images
-
         return images
 
     def transform_labels(self, labels, transformation, training=True):
+        if training:
+            labels = self._mix_up_labels(labels, transformation)
+        return labels
+
+    def _mix_up_labels(self, labels, transformation):
         mix_weight = transformation["mix_weight"]
         permutation_order = transformation["permutation_order"]
-
         labels_for_mix_up = self.backend.numpy.take(
             labels, permutation_order, axis=0
         )
-
         mix_weight = self.backend.numpy.reshape(mix_weight, [-1, 1])
-
         labels = mix_weight * labels + (1.0 - mix_weight) * labels_for_mix_up
-
         return labels
 
     def transform_bounding_boxes(
@@ -104,6 +106,11 @@ def transform_bounding_boxes(
         transformation,
         training=True,
     ):
+        if training:
+            return self._mix_up_bounding_boxes(bounding_boxes, transformation)
+        return bounding_boxes
+
+    def _mix_up_bounding_boxes(self, bounding_boxes, transformation):
         permutation_order = transformation["permutation_order"]
         boxes, classes = bounding_boxes["boxes"], bounding_boxes["classes"]
         boxes_for_mix_up = self.backend.numpy.take(boxes, permutation_order)
@@ -117,20 +124,23 @@ def transform_bounding_boxes(
     def transform_segmentation_masks(
         self, segmentation_masks, transformation, training=True
     ):
+        if training:
+            segmentation_masks = self._mix_up_segmentation_masks(
+                segmentation_masks, transformation
+            )
+        return segmentation_masks
+
+    def _mix_up_segmentation_masks(self, segmentation_masks, transformation):
         mix_weight = transformation["mix_weight"]
         permutation_order = transformation["permutation_order"]
-
         mix_weight = self.backend.numpy.reshape(mix_weight, [-1, 1, 1, 1])
-
         segmentation_masks_for_mix_up = self.backend.numpy.take(
             segmentation_masks, permutation_order
         )
-
         segmentation_masks = (
             mix_weight * segmentation_masks
             + (1.0 - mix_weight) * segmentation_masks_for_mix_up
         )
-
         return segmentation_masks
 
     def compute_output_shape(self, input_shape):

keras/src/layers/preprocessing/image_preprocessing/mix_up_test.py

@@ -21,6 +21,14 @@ def test_layer(self):
             run_training_check=not testing.tensorflow_uses_gpu(),
         )
 
+    def test_mix_up_inference(self):
+        seed = 3481
+        layer = layers.MixUp(alpha=0.2)
+        np.random.seed(seed)
+        inputs = np.random.randint(0, 255, size=(224, 224, 3))
+        output = layer(inputs, training=False)
+        self.assertAllClose(inputs, output)
+
     def test_mix_up_basic_functionality(self):
         image = np.random.random((64, 64, 3))
         mix_up_layer = layers.MixUp(alpha=1)

keras/src/layers/preprocessing/image_preprocessing/random_hue.py

@@ -107,16 +107,19 @@ def get_random_transformation(self, data, training=True, seed=None):
         return {"factor": invert * factor * 0.5}
 
     def transform_images(self, images, transformation=None, training=True):
+        if training:
+            images = self._apply_random_hue_to_images(images, transformation)
+        return images
+
+    def _apply_random_hue_to_images(self, images, transformation):
         images = transform_value_range(images, self.value_range, (0, 1))
         adjust_factors = transformation["factor"]
         adjust_factors = self.backend.cast(adjust_factors, images.dtype)
         adjust_factors = self.backend.numpy.expand_dims(adjust_factors, -1)
         adjust_factors = self.backend.numpy.expand_dims(adjust_factors, -1)
-
         images = self.backend.image.rgb_to_hsv(
             images, data_format=self.data_format
         )
-
         if self.data_format == "channels_first":
             h_channel = images[:, 0, :, :] + adjust_factors
             h_channel = self.backend.numpy.where(
@@ -142,10 +145,8 @@ def transform_images(self, images, transformation=None, training=True):
         images = self.backend.image.hsv_to_rgb(
             images, data_format=self.data_format
         )
-
         images = self.backend.numpy.clip(images, 0, 1)
         images = transform_value_range(images, (0, 1), self.value_range)
-
         return images
 
     def transform_labels(self, labels, transformation, training=True):

keras/src/layers/preprocessing/image_preprocessing/random_hue_test.py

@@ -23,6 +23,14 @@ def test_layer(self):
             expected_output_shape=(8, 3, 4, 3),
         )
 
+    def test_random_hue_inference(self):
+        seed = 3481
+        layer = layers.RandomHue(0.2, [0, 1.0])
+        np.random.seed(seed)
+        inputs = np.random.randint(0, 255, size=(224, 224, 3))
+        output = layer(inputs, training=False)
+        self.assertAllClose(inputs, output)
+
     def test_random_hue_value_range(self):
         image = keras.random.uniform(shape=(3, 3, 3), minval=0, maxval=1)