Add segmask support for random_flip (#775)

* Add segmask support for random_flip

* Set seed for test case

* Undo RRC changes

* Make shared _flip_image method

Author: ianstenbit

keras_cv/layers/preprocessing/random_flip.py

@@ -85,18 +85,7 @@ def augment_label(self, label, transformation, **kwargs):
         return label
 
     def augment_image(self, image, transformation, **kwargs):
-        flipped_output = tf.cond(
-            transformation["flip_horizontal"],
-            lambda: tf.image.flip_left_right(image),
-            lambda: image,
-        )
-        flipped_output = tf.cond(
-            transformation["flip_vertical"],
-            lambda: tf.image.flip_up_down(flipped_output),
-            lambda: flipped_output,
-        )
-        flipped_output.set_shape(image.shape)
-        return flipped_output
+        return RandomFlip._flip_image(image, transformation)
 
     def get_random_transformation(self, **kwargs):
         flip_horizontal = False
@@ -110,6 +99,20 @@ def get_random_transformation(self, **kwargs):
             "flip_vertical": tf.cast(flip_vertical, dtype=tf.bool),
         }
 
+    def _flip_image(image, transformation):
+        flipped_output = tf.cond(
+            transformation["flip_horizontal"],
+            lambda: tf.image.flip_left_right(image),
+            lambda: image,
+        )
+        flipped_output = tf.cond(
+            transformation["flip_vertical"],
+            lambda: tf.image.flip_up_down(flipped_output),
+            lambda: flipped_output,
+        )
+        flipped_output.set_shape(image.shape)
+        return flipped_output
+
     def _flip_bounding_boxes_horizontal(bounding_boxes):
         x1, x2, x3, x4, rest = tf.split(
             bounding_boxes, [1, 1, 1, 1, bounding_boxes.shape[-1] - 4], axis=-1
@@ -186,6 +189,11 @@ def augment_bounding_boxes(
         )
         return bounding_boxes
 
+    def augment_segmentation_mask(
+        self, segmentation_mask, transformation=None, **kwargs
+    ):
+        return RandomFlip._flip_image(segmentation_mask, transformation)
+
     def compute_output_shape(self, input_shape):
         return input_shape
 

keras_cv/layers/preprocessing/random_flip_test.py

@@ -134,3 +134,25 @@ def test_augment_bbox_batched_input(self):
         )
         expected_output = np.reshape(expected_output, (2, 2, 5))
         self.assertAllClose(expected_output, output["bounding_boxes"])
+
+    def test_augment_segmentation_mask(self):
+        np.random.seed(1337)
+        image = np.random.random((1, 20, 20, 3)).astype(np.float32)
+        mask = np.random.randint(2, size=(1, 20, 20, 1)).astype(np.float32)
+
+        input = {"images": image, "segmentation_masks": mask}
+
+        # Flip both vertically and horizontally
+        mock_random = [0.6, 0.6]
+        layer = RandomFlip()
+
+        with unittest.mock.patch.object(
+            layer._random_generator,
+            "random_uniform",
+            side_effect=mock_random,
+        ):
+            output = layer(input, training=True)
+
+        expected_mask = np.flip(np.flip(mask, axis=1), axis=2)
+
+        self.assertAllClose(expected_mask, output["segmentation_masks"])