Rename num_classes to classes across entire repo (#657)

* Rename num_classes to classes

* Fix find and replace mistake

* Fix accidental fstring removal

* Rename num_classes to classes for non-model uses

* More find-and-replace fixes

* One more f+r fix

* Fix non_max_supression

* Rename targets to class_predictions in non_max_supression

benchmarks/metrics/coco/mean_average_precision_bucket_performance.py

@@ -12,7 +12,7 @@
 from keras_cv.metrics import coco
 
 
-def produce_random_data(include_confidence=False, num_images=128, num_classes=20):
+def produce_random_data(include_confidence=False, num_images=128, classes=20):
     """Generates a fake list of bounding boxes for use in this test.
 
     Returns:
@@ -23,9 +23,9 @@ def produce_random_data(include_confidence=False, num_images=128, num_classes=20
     images = []
     for _ in range(num_images):
         num_boxes = math.floor(25 * random.uniform(0, 1))
-        classes = np.floor(np.random.rand(num_boxes, 1) * num_classes)
+        classes_in_image = np.floor(np.random.rand(num_boxes, 1) * classes)
         bboxes = np.random.rand(num_boxes, 4)
-        boxes = np.concatenate([bboxes, classes], axis=-1)
+        boxes = np.concatenate([bboxes, classes_in_image], axis=-1)
         if include_confidence:
             confidence = np.random.rand(num_boxes, 1)
             boxes = np.concatenate([boxes, confidence], axis=-1)

benchmarks/metrics/coco/mean_average_precision_performance.py

@@ -12,7 +12,7 @@
 from keras_cv.metrics import coco
 
 
-def produce_random_data(include_confidence=False, num_images=128, num_classes=20):
+def produce_random_data(include_confidence=False, num_images=128, classes=20):
     """Generates a fake list of bounding boxes for use in this test.
 
     Returns:
@@ -23,9 +23,9 @@ def produce_random_data(include_confidence=False, num_images=128, num_classes=20
     images = []
     for _ in range(num_images):
         num_boxes = math.floor(25 * random.uniform(0, 1))
-        classes = np.floor(np.random.rand(num_boxes, 1) * num_classes)
+        classes_in_image = np.floor(np.random.rand(num_boxes, 1) * classes)
         bboxes = np.random.rand(num_boxes, 4)
-        boxes = np.concatenate([bboxes, classes], axis=-1)
+        boxes = np.concatenate([bboxes, classes_in_image], axis=-1)
         if include_confidence:
             confidence = np.random.rand(num_boxes, 1)
             boxes = np.concatenate([boxes, confidence], axis=-1)

benchmarks/metrics/coco/recall_performance.py

@@ -12,7 +12,7 @@
 from keras_cv.metrics import coco
 
 
-def produce_random_data(include_confidence=False, num_images=128, num_classes=20):
+def produce_random_data(include_confidence=False, num_images=128, classes=20):
     """Generates a fake list of bounding boxes for use in this test.
 
     Returns:
@@ -23,9 +23,9 @@ def produce_random_data(include_confidence=False, num_images=128, num_classes=20
     images = []
     for _ in range(num_images):
         num_boxes = math.floor(25 * random.uniform(0, 1))
-        classes = np.floor(np.random.rand(num_boxes, 1) * num_classes)
+        classes_in_image = np.floor(np.random.rand(num_boxes, 1) * classes)
         bboxes = np.random.rand(num_boxes, 4)
-        boxes = np.concatenate([bboxes, classes], axis=-1)
+        boxes = np.concatenate([bboxes, classes_in_image], axis=-1)
         if include_confidence:
             confidence = np.random.rand(num_boxes, 1)
             boxes = np.concatenate([boxes, confidence], axis=-1)

examples/layers/preprocessing/classification/demo_utils.py

@@ -17,9 +17,9 @@
 import tensorflow_datasets as tfds
 
 
-def resize(image, label, img_size=(224, 224), num_classes=10):
+def resize(image, label, img_size=(224, 224), classes=10):
     image = tf.image.resize(image, img_size)
-    label = tf.one_hot(label, num_classes)
+    label = tf.one_hot(label, classes)
     return {"images": image, "labels": label}
 
 
@@ -32,11 +32,11 @@ def load_oxford_dataset(
     # Load dataset.
     data, ds_info = tfds.load(name, as_supervised=as_supervised, with_info=True)
     train_ds = data["train"]
-    num_classes = ds_info.features["label"].num_classes
+    classes = ds_info.features["label"].num_classes
 
     # Get tf dataset.
     train_ds = train_ds.map(
-        lambda x, y: resize(x, y, img_size=img_size, num_classes=num_classes)
+        lambda x, y: resize(x, y, img_size=img_size, classes=classes)
     ).batch(batch_size)
     return train_ds
 

examples/training/classification/densenet/imagenet/basic_training.py

@@ -61,7 +61,7 @@
 FLAGS = flags.FLAGS
 FLAGS(sys.argv)
 
-NUM_CLASSES = 1000
+CLASSES = 1000
 BATCH_SIZE = 256
 IMAGE_SIZE = (224, 224)
 EPOCHS = 250
@@ -94,7 +94,7 @@ def parse_imagenet_example(example):
 
     # Decode label
     label = tf.cast(tf.reshape(parsed[label_key], shape=()), dtype=tf.int32) - 1
-    label = tf.one_hot(label, NUM_CLASSES)
+    label = tf.one_hot(label, CLASSES)
     return image, label
 
 
@@ -164,7 +164,7 @@ def augment(img, label):
     model = DenseNet121(
         include_rescaling=True,
         include_top=True,
-        num_classes=NUM_CLASSES,
+        classes=CLASSES,
         input_shape=IMAGE_SIZE + (3,),
     )
 

keras_cv/layers/object_detection/non_max_suppression.py

@@ -29,7 +29,7 @@ class NonMaxSuppression(tf.keras.layers.Layer):
         - [Yolo paper](https://arxiv.org/pdf/1506.02640)
 
     Args:
-        num_classes: an integer representing the number of classes that a bounding
+        classes: an integer representing the number of classes that a bounding
             box can belong to.
         bounding_box_format: a case-insensitive string which is one of `"xyxy"`,
             `"rel_xyxy"`, `"xyWH"`, `"center_xyWH"`, `"yxyx"`, `"rel_yxyx"`. The
@@ -67,7 +67,7 @@ class NonMaxSuppression(tf.keras.layers.Layer):
     ], dtype = np.float32)
 
     nms = NonMaxSuppression(
-        num_classes=8,
+        classes=8,
         bounding_box_format="center_xyWH",
         iou_threshold=0.1
     )
@@ -78,7 +78,7 @@ class NonMaxSuppression(tf.keras.layers.Layer):
 
     def __init__(
         self,
-        num_classes,
+        classes,
         bounding_box_format,
         confidence_threshold=0.05,
         iou_threshold=0.5,
@@ -87,7 +87,7 @@ def __init__(
         **kwargs,
     ):
         super().__init__(**kwargs)
-        self.num_classes = num_classes
+        self.classes = classes
         self.bounding_box_format = bounding_box_format
         self.confidence_threshold = confidence_threshold
         self.iou_threshold = iou_threshold
@@ -112,11 +112,11 @@ def call(self, predictions, images=None):
 
         # preparing the predictions for TF NMS op
         boxes = tf.expand_dims(predictions[..., :4], axis=2)
-        classes = tf.cast(predictions[..., 4], tf.int32)
+        class_predictions = tf.cast(predictions[..., 4], tf.int32)
         scores = predictions[..., 5]
 
-        classes = tf.one_hot(classes, self.num_classes)
-        scores = tf.expand_dims(scores, axis=-1) * classes
+        class_predictions = tf.one_hot(class_predictions, self.classes)
+        scores = tf.expand_dims(scores, axis=-1) * class_predictions
 
         # applying the NMS operation
         nmsed_boxes = tf.image.combined_non_max_suppression(
@@ -181,7 +181,7 @@ def _encode_to_ragged(self, boxes, valid_detections):
 
     def get_config(self):
         config = {
-            "num_classes": self.num_classes,
+            "classes": self.classes,
             "bounding_box_format": self.bounding_box_format,
             "confidence_threshold": self.confidence_threshold,
             "iou_threshold": self.iou_threshold,

keras_cv/layers/object_detection/non_max_suppression_test.py

@@ -19,7 +19,7 @@
 
 class NonMaxSuppressionTest(tf.test.TestCase):
     def test_return_shapes(self):
-        layer = NonMaxSuppression(num_classes=4, bounding_box_format="xyWH")
+        layer = NonMaxSuppression(classes=4, bounding_box_format="xyWH")
         images = tf.ones((3, 480, 480, 3))
 
         boxes = tf.cast(tf.random.uniform((3, 5, 4), 0, 480, tf.int32), tf.float32)
@@ -32,7 +32,7 @@ def test_return_shapes(self):
         self.assertEqual(boxes.shape, [3, None, 6])
 
     def test_non_square_images(self):
-        layer = NonMaxSuppression(num_classes=4, bounding_box_format="xyxy")
+        layer = NonMaxSuppression(classes=4, bounding_box_format="xyxy")
 
         boxes = tf.cast(tf.random.uniform((2, 5, 4), 0, 480, tf.int32), tf.float32)
         classes = tf.cast(tf.random.uniform((2, 5, 1), 0, 4, tf.int32), tf.float32)
@@ -51,7 +51,7 @@ def test_non_square_images(self):
         self.assertEqual(boxes.shape, [2, None, 6])
 
     def test_different_channels(self):
-        layer = NonMaxSuppression(num_classes=4, bounding_box_format="xyWH")
+        layer = NonMaxSuppression(classes=4, bounding_box_format="xyWH")
         images = tf.ones((3, 480, 480, 5))
 
         boxes = tf.cast(tf.random.uniform((3, 5, 4), 0, 480, tf.int32), tf.float32)
@@ -66,7 +66,7 @@ def test_different_channels(self):
     def test_in_a_model(self):
         input1 = tf.keras.layers.Input([5, 6])
         input2 = tf.keras.layers.Input([480, 480, 3])
-        layer = NonMaxSuppression(num_classes=4, bounding_box_format="xyWH")
+        layer = NonMaxSuppression(classes=4, bounding_box_format="xyWH")
         outputs = layer(input1, input2)
 
         model = tf.keras.models.Model(inputs=[input1, input2], outputs=outputs)
@@ -83,7 +83,7 @@ def test_in_a_model(self):
         self.assertEqual(boxes.shape, [3, None, 6])
 
     def test_without_images(self):
-        layer = NonMaxSuppression(num_classes=4, bounding_box_format="xyWH")
+        layer = NonMaxSuppression(classes=4, bounding_box_format="xyWH")
 
         boxes = tf.cast(tf.random.uniform((3, 5, 4), 0, 480, tf.int32), tf.float32)
         classes = tf.cast(tf.random.uniform((3, 5, 1), 0, 4, tf.int32), tf.float32)

keras_cv/layers/preprocessing/cut_mix_test.py

@@ -15,7 +15,7 @@
 
 from keras_cv.layers.preprocessing.cut_mix import CutMix
 
-NUM_CLASSES = 10
+classes = 10
 
 
 class CutMixTest(tf.test.TestCase):
@@ -24,7 +24,7 @@ def test_return_shapes(self):
         # randomly sample labels
         ys = tf.random.categorical(tf.math.log([[0.5, 0.5]]), 2)
         ys = tf.squeeze(ys)
-        ys = tf.one_hot(ys, NUM_CLASSES)
+        ys = tf.one_hot(ys, classes)
 
         layer = CutMix(seed=1)
         outputs = layer({"images": xs, "labels": ys})

keras_cv/layers/preprocessing/fourier_mix_test.py

@@ -15,7 +15,7 @@
 
 from keras_cv.layers.preprocessing.fourier_mix import FourierMix
 
-NUM_CLASSES = 10
+classes = 10
 
 
 class FourierMixTest(tf.test.TestCase):
@@ -24,7 +24,7 @@ def test_return_shapes(self):
         # randomly sample labels
         ys = tf.random.categorical(tf.math.log([[0.5, 0.5]]), 2)
         ys = tf.squeeze(ys)
-        ys = tf.one_hot(ys, NUM_CLASSES)
+        ys = tf.one_hot(ys, classes)
 
         layer = FourierMix()
         outputs = layer({"images": xs, "labels": ys})

keras_cv/layers/preprocessing/mix_up_test.py

@@ -15,7 +15,7 @@
 
 from keras_cv.layers.preprocessing.mix_up import MixUp
 
-NUM_CLASSES = 10
+classes = 10
 
 
 class MixUpTest(tf.test.TestCase):
@@ -24,7 +24,7 @@ def test_return_shapes(self):
         # randomly sample labels
         ys_labels = tf.random.categorical(tf.math.log([[0.5, 0.5]]), 2)
         ys_labels = tf.squeeze(ys_labels)
-        ys_labels = tf.one_hot(ys_labels, NUM_CLASSES)
+        ys_labels = tf.one_hot(ys_labels, classes)
 
         # randomly sample bounding boxes
         ys_bounding_boxes = tf.random.uniform((2, 3, 5), 0, 1)

keras_cv/layers/preprocessing/random_shear_test.py

@@ -15,7 +15,7 @@
 
 from keras_cv.layers import preprocessing
 
-NUM_CLASSES = 10
+classes = 10
 
 
 class RandomShearTest(tf.test.TestCase):
@@ -45,7 +45,7 @@ def test_return_shapes(self):
         # randomly sample labels
         ys_labels = tf.random.categorical(tf.math.log([[0.5, 0.5]]), 2)
         ys_labels = tf.squeeze(ys_labels)
-        ys_labels = tf.one_hot(ys_labels, NUM_CLASSES)
+        ys_labels = tf.one_hot(ys_labels, classes)
 
         # randomly sample bounding boxes
         ys_bounding_boxes = tf.random.uniform((2, 3, 7), 0, 1)

keras_cv/layers/serialization_test.py

@@ -244,7 +244,7 @@ class SerializationTest(tf.test.TestCase, parameterized.TestCase):
             "NonMaxSuppression",
             object_detection.NonMaxSuppression,
             {
-                "num_classes": 5,
+                "classes": 5,
                 "bounding_box_format": "xyxy",
                 "confidence_threshold": 0.5,
                 "iou_threshold": 0.5,

keras_cv/metrics/coco/mean_average_precision.py

@@ -31,7 +31,7 @@ class COCOMeanAveragePrecision(tf.keras.metrics.Metric):
     Args:
         class_ids: The class IDs to evaluate the metric for.  To evaluate for
             all classes in over a set of sequentially labelled classes, pass
-            `range(num_classes)`.
+            `range(classes)`.
         bounding_box_format: Format of the incoming bounding boxes.  Supported values
             are "xywh", "center_xywh", "xyxy".
         iou_thresholds: IoU thresholds over which to evaluate the recall.  Must

keras_cv/metrics/coco/recall.py

@@ -33,7 +33,7 @@ class COCORecall(keras.metrics.Metric):
     Args:
         class_ids: The class IDs to evaluate the metric for.  To evaluate for
             all classes in over a set of sequentially labelled classes, pass
-            `range(num_classes)`.
+            `range(classes)`.
         bounding_box_format: Format of the incoming bounding boxes.  Supported values
             are "xywh", "center_xywh", "xyxy".
         iou_thresholds: IoU thresholds over which to evaluate the recall.  Must

keras_cv/models/csp_darknet.py

@@ -38,7 +38,7 @@ def CSPDarkNet(
     depth_multiplier=1.0,
     width_multiplier=1.0,
     use_depthwise=False,
-    num_classes=None,
+    classes=None,
     weights=None,
     input_shape=(None, None, 3),
     input_tensor=None,
@@ -65,14 +65,14 @@ def CSPDarkNet(
         include_rescaling: whether or not to Rescale the inputs.If set to True,
             inputs will be passed through a `Rescaling(1/255.0)` layer.
         include_top: whether to include the fully-connected layer at the top of
-            the network.  If provided, `num_classes` must be provided.
+            the network.  If provided, `classes` must be provided.
         depth_multiplier: A float value used to calculate the base depth of the model
             this changes based the detection model being used. Defaults to 1.0.
         width_multiplier: A float value used to calculate the base width of the model
             this changes based the detection model being used. Defaults to 1.0.
         use_depthwise: a boolean value used to decide whether a depthwise conv block
             should be used over a regular darknet block. Defaults to False
-        num_classes: optional number of classes to classify images into, only to be
+        classes: optional number of classes to classify images into, only to be
             specified if `include_top` is True, and if no `weights` argument is
             specified.
         weights: one of `None` (random initialization), or a pretrained weight
@@ -102,10 +102,10 @@ def CSPDarkNet(
             f"weights file to be loaded. Weights file not found at location: {weights}"
         )
 
-    if include_top and not num_classes:
+    if include_top and not classes:
         raise ValueError(
-            "If `include_top` is True, you should specify `num_classes`. Received: "
-            f"num_classes={num_classes}"
+            "If `include_top` is True, you should specify `classes`. Received: "
+            f"classes={classes}"
         )
 
     ConvBlock = DarknetConvBlockDepthwise if use_depthwise else DarknetConvBlock
@@ -165,9 +165,9 @@ def CSPDarkNet(
 
     if include_top:
         x = layers.GlobalAveragePooling2D(name="avg_pool")(x)
-        x = layers.Dense(
-            num_classes, activation=classifier_activation, name="predictions"
-        )(x)
+        x = layers.Dense(classes, activation=classifier_activation, name="predictions")(
+            x
+        )
     elif pooling == "avg":
         x = layers.GlobalAveragePooling2D(name="avg_pool")(x)
     elif pooling == "max":