Migrating Fully Convolutional Network example to Keras 3 (keras-team#…

…1691)

* Migrated fully_convolutional_network example to Keras 3

* Initialised seed to integer values

examples/vision/fully_convolutional_network.py

@@ -2,7 +2,7 @@
 Title: Image Segmentation using Composable Fully-Convolutional Networks
 Author: [Suvaditya Mukherjee](https://twitter.com/halcyonrayes)
 Date created: 2023/06/16
-Last modified: 2023/06/16
+Last modified: 2023/12/25
 Description: Using the Fully-Convolutional Network for Image Segmentation.
 Accelerator: GPU
 """
@@ -48,15 +48,15 @@
 ## Setup Imports
 """
 
+import os
+os.environ["KERAS_BACKEND"] = "tensorflow"
+import keras
+from keras import ops
 import tensorflow as tf
-from tensorflow import keras
 import matplotlib.pyplot as plt
 import tensorflow_datasets as tfds
 import numpy as np
 
-keras.utils.set_random_seed(27)
-tf.random.set_seed(27)
-
 AUTOTUNE = tf.data.AUTOTUNE
 
 """
@@ -147,10 +147,10 @@ def unpack_resize_data(section):
 # for Matplotlib visualization.
 
 images, masks = next(iter(test_ds))
-random_idx = tf.random.uniform([], minval=0, maxval=BATCH_SIZE, dtype=tf.int32)
+random_idx = keras.random.uniform([], minval=0, maxval=BATCH_SIZE, seed=10)
 
-test_image = images[random_idx].numpy().astype("float")
-test_mask = masks[random_idx].numpy().astype("float")
+test_image = images[int(random_idx)].numpy().astype("float")
+test_mask = masks[int(random_idx)].numpy().astype("float")
 
 # Overlay segmentation mask on top of image.
 fig, ax = plt.subplots(nrows=1, ncols=2, figsize=(10, 5))
@@ -259,7 +259,7 @@ def preprocess_data(image, segmentation_mask):
         activation="relu",
         padding="same",
         use_bias=False,
-        kernel_initializer=tf.constant_initializer(1.0),
+        kernel_initializer=keras.initializers.Constant(1.0),
     )
     dense_convs.append(dense_conv)
     dropout_layer = keras.layers.Dropout(0.5)
@@ -556,13 +556,13 @@ def preprocess_data(image, segmentation_mask):
 """
 
 images, masks = next(iter(test_ds))
-random_idx = tf.random.uniform([], minval=0, maxval=BATCH_SIZE, dtype=tf.int32)
+random_idx = keras.random.uniform([], minval=0, maxval=BATCH_SIZE,seed=10)
 
 # Get random test image and mask
-test_image = images[random_idx].numpy().astype("float")
-test_mask = masks[random_idx].numpy().astype("float")
+test_image = images[int(random_idx)].numpy().astype("float")
+test_mask = masks[int(random_idx)].numpy().astype("float")
 
-pred_image = tf.expand_dims(test_image, axis=0)
+pred_image = ops.expand_dims(test_image, axis=0)
 pred_image = keras.applications.vgg19.preprocess_input(pred_image)
 
 # Perform inference on FCN-32S

examples/vision/ipynb/fully_convolutional_network.ipynb

@@ -11,7 +11,7 @@
     "\n",
     "**Author:** [Suvaditya Mukherjee](https://twitter.com/halcyonrayes)<br>\n",
     "**Date created:** 2023/06/16<br>\n",
-    "**Last modified:** 2023/06/16<br>\n",
+    "**Last modified:** 2023/12/25<br>\n",
     "**Description:** Using the Fully-Convolutional Network for Image Segmentation."
    ]
   },
@@ -75,15 +75,15 @@
    },
    "outputs": [],
    "source": [
+    "import os\n",
+    "os.environ[\"KERAS_BACKEND\"] = \"tensorflow\"\n",
+    "import keras\n",
+    "from keras import ops\n",
     "import tensorflow as tf\n",
-    "from tensorflow import keras\n",
     "import matplotlib.pyplot as plt\n",
     "import tensorflow_datasets as tfds\n",
     "import numpy as np\n",
     "\n",
-    "keras.utils.set_random_seed(27)\n",
-    "tf.random.set_seed(27)\n",
-    "\n",
     "AUTOTUNE = tf.data.AUTOTUNE"
    ]
   },
@@ -235,10 +235,10 @@
     "# for Matplotlib visualization.\n",
     "\n",
     "images, masks = next(iter(test_ds))\n",
-    "random_idx = tf.random.uniform([], minval=0, maxval=BATCH_SIZE, dtype=tf.int32)\n",
+    "random_idx = keras.random.uniform([], minval=0, maxval=BATCH_SIZE, seed=10)\n",
     "\n",
-    "test_image = images[random_idx].numpy().astype(\"float\")\n",
-    "test_mask = masks[random_idx].numpy().astype(\"float\")\n",
+    "test_image = images[int(random_idx)].numpy().astype(\"float\")\n",
+    "test_mask = masks[int(random_idx)].numpy().astype(\"float\")\n",
     "\n",
     "# Overlay segmentation mask on top of image.\n",
     "fig, ax = plt.subplots(nrows=1, ncols=2, figsize=(10, 5))\n",
@@ -384,7 +384,7 @@
     "        activation=\"relu\",\n",
     "        padding=\"same\",\n",
     "        use_bias=False,\n",
-    "        kernel_initializer=tf.constant_initializer(1.0),\n",
+    "        kernel_initializer=keras.initializers.Constant(1.0),\n",
     "    )\n",
     "    dense_convs.append(dense_conv)\n",
     "    dropout_layer = keras.layers.Dropout(0.5)\n",
@@ -829,13 +829,13 @@
    "outputs": [],
    "source": [
     "images, masks = next(iter(test_ds))\n",
-    "random_idx = tf.random.uniform([], minval=0, maxval=BATCH_SIZE, dtype=tf.int32)\n",
+    "random_idx = keras.random.uniform([], minval=0, maxval=BATCH_SIZE,seed=10)\n",
     "\n",
     "# Get random test image and mask\n",
-    "test_image = images[random_idx].numpy().astype(\"float\")\n",
-    "test_mask = masks[random_idx].numpy().astype(\"float\")\n",
+    "test_image = images[int(random_idx)].numpy().astype(\"float\")\n",
+    "test_mask = masks[int(random_idx)].numpy().astype(\"float\")\n",
     "\n",
-    "pred_image = tf.expand_dims(test_image, axis=0)\n",
+    "pred_image = ops.expand_dims(test_image, axis=0)\n",
     "pred_image = keras.applications.vgg19.preprocess_input(pred_image)\n",
     "\n",
     "# Perform inference on FCN-32S\n",

examples/vision/md/fully_convolutional_network.md

@@ -2,7 +2,7 @@
 
 **Author:** [Suvaditya Mukherjee](https://twitter.com/halcyonrayes)<br>
 **Date created:** 2023/06/16<br>
-**Last modified:** 2023/06/16<br>
+**Last modified:** 2023/12/25<br>
 **Description:** Using the Fully-Convolutional Network for Image Segmentation.
 
 
@@ -50,15 +50,15 @@ or a [PyImageSearch Blog on Semantic Segmentation](https://pyimagesearch.com/201
 
 
 ```python
+import os
+os.environ["KERAS_BACKEND"] = "tensorflow"
+import keras
+from keras import ops
 import tensorflow as tf
-from tensorflow import keras
 import matplotlib.pyplot as plt
 import tensorflow_datasets as tfds
 import numpy as np
 
-keras.utils.set_random_seed(27)
-tf.random.set_seed(27)
-
 AUTOTUNE = tf.data.AUTOTUNE
 ```
 
@@ -165,10 +165,10 @@ which makes the image and mask size same.
 # for Matplotlib visualization.
 
 images, masks = next(iter(test_ds))
-random_idx = tf.random.uniform([], minval=0, maxval=BATCH_SIZE, dtype=tf.int32)
+random_idx = keras.random.uniform([], minval=0, maxval=BATCH_SIZE, seed=10)
 
-test_image = images[random_idx].numpy().astype("float")
-test_mask = masks[random_idx].numpy().astype("float")
+test_image = images[int(random_idx)].numpy().astype("float")
+test_mask = masks[int(random_idx)].numpy().astype("float")
 
 # Overlay segmentation mask on top of image.
 fig, ax = plt.subplots(nrows=1, ncols=2, figsize=(10, 5))
@@ -286,7 +286,7 @@ for filter_idx in range(len(units)):
         activation="relu",
         padding="same",
         use_bias=False,
-        kernel_initializer=tf.constant_initializer(1.0),
+        kernel_initializer=keras.initializers.Constant(1.0),,
     )
     dense_convs.append(dense_conv)
     dropout_layer = keras.layers.Dropout(0.5)
@@ -972,13 +972,13 @@ Note: For better results, the model must be trained for a higher number of epoch
 
 ```python
 images, masks = next(iter(test_ds))
-random_idx = tf.random.uniform([], minval=0, maxval=BATCH_SIZE, dtype=tf.int32)
+random_idx = keras.random.uniform([], minval=0, maxval=BATCH_SIZE,seed=10)
 
 # Get random test image and mask
-test_image = images[random_idx].numpy().astype("float")
-test_mask = masks[random_idx].numpy().astype("float")
+test_image = images[int(random_idx)].numpy().astype("float")
+test_mask = masks[int(random_idx)].numpy().astype("float")
 
-pred_image = tf.expand_dims(test_image, axis=0)
+pred_image = ops.expand_dims(test_image, axis=0)
 pred_image = keras.applications.vgg19.preprocess_input(pred_image)
 
 # Perform inference on FCN-32S