Updates Keras CV - Classification guide for Keras 3

guides/ipynb/keras_cv/classification_with_keras_cv.ipynb

@@ -33,6 +33,11 @@
     "- Fine-tuning a pretrained backbone\n",
     "- Training a image classifier from scratch\n",
     "\n",
+    "KerasCV uses Keras 3 to work with any of TensorFlow, PyTorch or Jax. In the\n",
+    "guide below, we will use the `jax` backend. This guide runs in\n",
+    "TensorFlow or PyTorch backends with zero changes, simply update the\n",
+    "`KERAS_BACKEND` below.\n",
+    "\n",
     "We use Professor Keras, the official Keras mascot, as a\n",
     "visual reference for the complexity of the material:\n",
     "\n",
@@ -47,18 +52,38 @@
    },
    "outputs": [],
    "source": [
+    "!pip install -q --upgrade keras-cv\n",
+    "!pip install -q --upgrade keras  # Upgrade to Keras 3."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 0,
+   "metadata": {
+    "colab_type": "code"
+   },
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "\n",
+    "os.environ[\"KERAS_BACKEND\"] = \"jax\"  # @param [\"tensorflow\", \"jax\", \"torch\"]\n",
     "\n",
     "import json\n",
     "import math\n",
-    "import keras_cv\n",
-    "import tensorflow as tf\n",
-    "import tensorflow_datasets as tfds\n",
+    "import numpy as np\n",
+    "\n",
     "import keras\n",
     "from keras import losses\n",
-    "import numpy as np\n",
+    "from keras import ops\n",
     "from keras import optimizers\n",
-    "from tensorflow.keras.optimizers import schedules\n",
+    "from keras.optimizers import schedules\n",
     "from keras import metrics\n",
+    "\n",
+    "import keras_cv\n",
+    "\n",
+    "# Import tensorflow for `tf.data` and its preprocessing functions\n",
+    "import tensorflow as tf\n",
+    "import tensorflow_datasets as tfds\n",
     ""
    ]
   },
@@ -126,11 +151,11 @@
    },
    "outputs": [],
    "source": [
-    "filepath = tf.keras.utils.get_file(origin=\"https://i.imgur.com/9i63gLN.jpg\")\n",
+    "filepath = keras.utils.get_file(origin=\"https://i.imgur.com/9i63gLN.jpg\")\n",
     "image = keras.utils.load_img(filepath)\n",
     "image = np.array(image)\n",
     "keras_cv.visualization.plot_image_gallery(\n",
-    "    [image], rows=1, cols=1, value_range=(0, 255), show=True, scale=4\n",
+    "    np.array([image]), rows=1, cols=1, value_range=(0, 255), show=True, scale=4\n",
     ")"
    ]
   },
@@ -325,7 +350,7 @@
     ")\n",
     "model.compile(\n",
     "    loss=\"categorical_crossentropy\",\n",
-    "    optimizer=tf.optimizers.SGD(learning_rate=0.01),\n",
+    "    optimizer=keras.optimizers.SGD(learning_rate=0.01),\n",
     "    metrics=[\"accuracy\"],\n",
     ")"
    ]
@@ -618,8 +643,10 @@
    "source": [
     "rand_augment = keras_cv.layers.RandAugment(\n",
     "    augmentations_per_image=3,\n",
-    "    magnitude=0.3,\n",
     "    value_range=(0, 255),\n",
+    "    magnitude=0.3,\n",
+    "    magnitude_stddev=0.2,\n",
+    "    rate=1.0,\n",
     ")\n",
     "augmenters += [rand_augment]\n",
     "\n",
@@ -793,8 +820,18 @@
    },
    "outputs": [],
    "source": [
-    "augmenter = keras.Sequential(augmenters)\n",
-    "train_ds = train_ds.map(augmenter, num_parallel_calls=tf.data.AUTOTUNE)\n",
+    "\n",
+    "def create_augmenter_fn(augmenters):\n",
+    "    def augmenter_fn(inputs):\n",
+    "        for augmenter in augmenters:\n",
+    "            inputs = augmenter(inputs)\n",
+    "        return inputs\n",
+    "\n",
+    "    return augmenter_fn\n",
+    "\n",
+    "\n",
+    "augmenter_fn = create_augmenter_fn(augmenters)\n",
+    "train_ds = train_ds.map(augmenter_fn, num_parallel_calls=tf.data.AUTOTUNE)\n",
     "\n",
     "image_batch = next(iter(train_ds.take(1)))[\"images\"]\n",
     "keras_cv.visualization.plot_image_gallery(\n",
@@ -913,23 +950,26 @@
     "        * target_lr\n",
     "        * (\n",
     "            1\n",
-    "            + tf.cos(\n",
-    "                tf.constant(math.pi)\n",
-    "                * tf.cast(global_step - warmup_steps - hold, tf.float32)\n",
-    "                / float(total_steps - warmup_steps - hold)\n",
+    "            + ops.cos(\n",
+    "                math.pi\n",
+    "                * ops.convert_to_tensor(\n",
+    "                    global_step - warmup_steps - hold, dtype=\"float32\"\n",
+    "                )\n",
+    "                / ops.convert_to_tensor(\n",
+    "                    total_steps - warmup_steps - hold, dtype=\"float32\"\n",
+    "                )\n",
     "            )\n",
     "        )\n",
     "    )\n",
     "\n",
-    "    warmup_lr = tf.cast(target_lr * (global_step / warmup_steps), tf.float32)\n",
-    "    target_lr = tf.cast(target_lr, tf.float32)\n",
+    "    warmup_lr = target_lr * (global_step / warmup_steps)\n",
     "\n",
     "    if hold > 0:\n",
-    "        learning_rate = tf.where(\n",
+    "        learning_rate = ops.where(\n",
     "            global_step > warmup_steps + hold, learning_rate, target_lr\n",
     "        )\n",
     "\n",
-    "    learning_rate = tf.where(global_step < warmup_steps, warmup_lr, learning_rate)\n",
+    "    learning_rate = ops.where(global_step < warmup_steps, warmup_lr, learning_rate)\n",
     "    return learning_rate\n",
     "\n",
     "\n",
@@ -952,7 +992,7 @@
     "            hold=self.hold,\n",
     "        )\n",
     "\n",
-    "        return tf.where(step > self.total_steps, 0.0, lr, name=\"learning_rate\")\n",
+    "        return ops.where(step > self.total_steps, 0.0, lr)\n",
     ""
    ]
   },
@@ -989,7 +1029,7 @@
     "    hold=hold_steps,\n",
     ")\n",
     "optimizer = optimizers.SGD(\n",
-    "    decay=5e-4,\n",
+    "    weight_decay=5e-4,\n",
     "    learning_rate=schedule,\n",
     "    momentum=0.9,\n",
     ")"
@@ -1015,7 +1055,7 @@
    },
    "outputs": [],
    "source": [
-    "backbone = keras_cv.models.EfficientNetV2B1Backbone()\n",
+    "backbone = keras_cv.models.EfficientNetV2B0Backbone()\n",
     "model = keras.Sequential(\n",
     "    [\n",
     "        backbone,\n",
@@ -1166,4 +1206,4 @@
  },
  "nbformat": 4,
  "nbformat_minor": 0
-}
+}

guides/keras_cv/classification_with_keras_cv.py

@@ -21,25 +21,43 @@
 - Fine-tuning a pretrained backbone
 - Training a image classifier from scratch
 
+KerasCV uses Keras 3 to work with any of TensorFlow, PyTorch or Jax. In the
+guide below, we will use the `jax` backend. This guide runs in
+TensorFlow or PyTorch backends with zero changes, simply update the
+`KERAS_BACKEND` below.
+
 We use Professor Keras, the official Keras mascot, as a
 visual reference for the complexity of the material:
 
 ![](https://storage.googleapis.com/keras-nlp/getting_started_guide/prof_keras_evolution.png)
 """
 
+"""shell
+pip install -q --upgrade keras-cv
+pip install -q --upgrade keras  # Upgrade to Keras 3.
+"""
+
+import os
+
+os.environ["KERAS_BACKEND"] = "jax"  # @param ["tensorflow", "jax", "torch"]
 
 import json
 import math
-import keras_cv
-import tensorflow as tf
-import tensorflow_datasets as tfds
+import numpy as np
+
 import keras
 from keras import losses
-import numpy as np
+from keras import ops
 from keras import optimizers
-from tensorflow.keras.optimizers import schedules
+from keras.optimizers import schedules
 from keras import metrics
 
+import keras_cv
+
+# Import tensorflow for `tf.data` and its preprocessing functions
+import tensorflow as tf
+import tensorflow_datasets as tfds
+
 
 """
 ## Inference with a pretrained classifier
@@ -77,11 +95,11 @@
 Now that our classifier is built, let's apply it to this cute cat picture!
 """
 
-filepath = tf.keras.utils.get_file(origin="https://i.imgur.com/9i63gLN.jpg")
+filepath = keras.utils.get_file(origin="https://i.imgur.com/9i63gLN.jpg")
 image = keras.utils.load_img(filepath)
 image = np.array(image)
 keras_cv.visualization.plot_image_gallery(
-    [image], rows=1, cols=1, value_range=(0, 255), show=True, scale=4
+    np.array([image]), rows=1, cols=1, value_range=(0, 255), show=True, scale=4
 )
 
 """
@@ -187,7 +205,7 @@ def preprocess_inputs(image, label):
 )
 model.compile(
     loss="categorical_crossentropy",
-    optimizer=tf.optimizers.SGD(learning_rate=0.01),
+    optimizer=keras.optimizers.SGD(learning_rate=0.01),
     metrics=["accuracy"],
 )
 
@@ -371,8 +389,10 @@ def package_inputs(image, label):
 """
 rand_augment = keras_cv.layers.RandAugment(
     augmentations_per_image=3,
-    magnitude=0.3,
     value_range=(0, 255),
+    magnitude=0.3,
+    magnitude_stddev=0.2,
+    rate=1.0,
 )
 augmenters += [rand_augment]
 
@@ -471,8 +491,18 @@ def package_inputs(image, label):
 Now let's apply our final augmenter to the training data:
 """
 
-augmenter = keras.Sequential(augmenters)
-train_ds = train_ds.map(augmenter, num_parallel_calls=tf.data.AUTOTUNE)
+
+def create_augmenter_fn(augmenters):
+    def augmenter_fn(inputs):
+        for augmenter in augmenters:
+            inputs = augmenter(inputs)
+        return inputs
+
+    return augmenter_fn
+
+
+augmenter_fn = create_augmenter_fn(augmenters)
+train_ds = train_ds.map(augmenter_fn, num_parallel_calls=tf.data.AUTOTUNE)
 
 image_batch = next(iter(train_ds.take(1)))["images"]
 keras_cv.visualization.plot_image_gallery(
@@ -548,23 +578,26 @@ def lr_warmup_cosine_decay(
         * target_lr
         * (
             1
-            + tf.cos(
-                tf.constant(math.pi)
-                * tf.cast(global_step - warmup_steps - hold, tf.float32)
-                / float(total_steps - warmup_steps - hold)
+            + ops.cos(
+                math.pi
+                * ops.convert_to_tensor(
+                    global_step - warmup_steps - hold, dtype="float32"
+                )
+                / ops.convert_to_tensor(
+                    total_steps - warmup_steps - hold, dtype="float32"
+                )
             )
         )
     )
 
-    warmup_lr = tf.cast(target_lr * (global_step / warmup_steps), tf.float32)
-    target_lr = tf.cast(target_lr, tf.float32)
+    warmup_lr = target_lr * (global_step / warmup_steps)
 
     if hold > 0:
-        learning_rate = tf.where(
+        learning_rate = ops.where(
             global_step > warmup_steps + hold, learning_rate, target_lr
         )
 
-    learning_rate = tf.where(global_step < warmup_steps, warmup_lr, learning_rate)
+    learning_rate = ops.where(global_step < warmup_steps, warmup_lr, learning_rate)
     return learning_rate
 
 
@@ -587,7 +620,7 @@ def __call__(self, step):
             hold=self.hold,
         )
 
-        return tf.where(step > self.total_steps, 0.0, lr, name="learning_rate")
+        return ops.where(step > self.total_steps, 0.0, lr)
 
 
 """
@@ -610,7 +643,7 @@ def __call__(self, step):
     hold=hold_steps,
 )
 optimizer = optimizers.SGD(
-    decay=5e-4,
+    weight_decay=5e-4,
     learning_rate=schedule,
     momentum=0.9,
 )
@@ -621,7 +654,7 @@ def __call__(self, step):
 Note that this preset does not come with any pretrained weights.
 """
 
-backbone = keras_cv.models.EfficientNetV2B1Backbone()
+backbone = keras_cv.models.EfficientNetV2B0Backbone()
 model = keras.Sequential(
     [
         backbone,