Update to Keras2 api

Author: d4nst

test/test_mnist.ipynb

@@ -95,14 +95,15 @@
     }
    ],
    "source": [
+    "batch_size = 128\n",
     "out = model.evaluate_generator(\n",
     "    RotNetDataGenerator(\n",
     "        X_test,\n",
-    "        batch_size=128,\n",
+    "        batch_size=batch_size,\n",
     "        preprocess_func=binarize_images,\n",
     "        shuffle=True\n",
     "    ), \n",
-    "    val_samples=len(y_test)\n",
+    "    steps=len(y_test) / batch_size\n",
     ")\n",
     "\n",
     "print('Test loss:', out[0])\n",

test/test_street_view.ipynb

@@ -103,17 +103,18 @@
     }
    ],
    "source": [
+    "batch_size = 64\n",
     "out = model.evaluate_generator(\n",
     "    RotNetDataGenerator(\n",
     "        test_filenames,\n",
     "        input_shape=(224, 224, 3),\n",
-    "        batch_size=64,\n",
+    "        batch_size=batch_size,\n",
     "        preprocess_func=preprocess_input,\n",
     "        crop_center=True,\n",
     "        crop_largest_rect=True,\n",
     "        shuffle=True\n",
     "    ),\n",
-    "    val_samples=len(test_filenames)\n",
+    "    steps=len(test_filenames) / batch_size\n",
     ")\n",
     "\n",
     "print('Test loss:', out[0])\n",

train/train_mnist.py

@@ -6,13 +6,11 @@
 from keras.callbacks import ModelCheckpoint, EarlyStopping, TensorBoard
 from keras.datasets import mnist
 from keras.layers import Dense, Dropout, Flatten, Input
-from keras.layers import Convolution2D, MaxPooling2D
+from keras.layers import Conv2D, MaxPooling2D
 from keras.models import Model
 
 sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 from utils import angle_error, RotNetDataGenerator, binarize_images
-
-
 # we don't need the labels indicating the digit value, so we only load the images
 (X_train, _), (X_test, _) = mnist.load_data()
 
@@ -38,18 +36,16 @@
 
 # model definition
 input = Input(shape=(img_rows, img_cols, img_channels))
-x = Convolution2D(nb_filters, kernel_size[0], kernel_size[1],
-                  activation='relu')(input)
-x = Convolution2D(nb_filters, kernel_size[0], kernel_size[1],
-                  activation='relu')(x)
+x = Conv2D(nb_filters, kernel_size, activation='relu')(input)
+x = Conv2D(nb_filters, kernel_size, activation='relu')(x)
 x = MaxPooling2D(pool_size=(2, 2))(x)
 x = Dropout(0.25)(x)
 x = Flatten()(x)
 x = Dense(128, activation='relu')(x)
 x = Dropout(0.25)(x)
 x = Dense(nb_classes, activation='softmax')(x)
 
-model = Model(input=input, output=x)
+model = Model(inputs=input, outputs=x)
 
 model.summary()
 
@@ -82,14 +78,14 @@
         preprocess_func=binarize_images,
         shuffle=True
     ),
-    samples_per_epoch=nb_train_samples,
-    nb_epoch=nb_epoch,
+    steps_per_epoch=nb_train_samples / batch_size,
+    epochs=nb_epoch,
     validation_data=RotNetDataGenerator(
         X_test,
         batch_size=batch_size,
         preprocess_func=binarize_images
     ),
-    nb_val_samples=nb_test_samples,
+    validation_steps=nb_test_samples / batch_size,
     verbose=1,
     callbacks=[checkpointer, early_stopping, tensorboard]
 )

train/train_mnist_regression.py

@@ -6,7 +6,7 @@
 from keras.callbacks import ModelCheckpoint, EarlyStopping, TensorBoard
 from keras.datasets import mnist
 from keras.layers import Dense, Dropout, Flatten, Input
-from keras.layers import Convolution2D, MaxPooling2D
+from keras.layers import Conv2D, MaxPooling2D
 from keras.models import Model
 
 sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
@@ -35,22 +35,16 @@
 
 # model definition
 input = Input(shape=(img_rows, img_cols, img_channels))
-x = Convolution2D(nb_filters, kernel_size[0], kernel_size[1],
-                  border_mode='valid',
-                  input_shape=input_shape,
-                  activation='relu')(input)
-x = Convolution2D(nb_filters, kernel_size[0], kernel_size[1],
-                  border_mode='valid',
-                  input_shape=input_shape,
-                  activation='relu')(x)
+x = Conv2D(nb_filters, kernel_size, activation='relu')(input)
+x = Conv2D(nb_filters, kernel_size, activation='relu')(x)
 x = MaxPooling2D(pool_size=(2, 2))(x)
 x = Dropout(0.25)(x)
 x = Flatten()(x)
 x = Dense(128, activation='relu')(x)
 x = Dropout(0.25)(x)
 x = Dense(1, activation='sigmoid')(x)
 
-model = Model(input=input, output=x)
+model = Model(inputs=input, outputs=x)
 
 model.summary()
 
@@ -84,15 +78,15 @@
         preprocess_func=binarize_images,
         shuffle=True
     ),
-    samples_per_epoch=nb_train_samples,
-    nb_epoch=nb_epoch,
+    steps_per_epoch=nb_train_samples / batch_size,
+    epochs=nb_epoch,
     validation_data=RotNetDataGenerator(
         X_test,
         one_hot=False,
         preprocess_func=binarize_images,
         batch_size=batch_size
     ),
-    nb_val_samples=nb_test_samples,
+    validation_steps=nb_test_samples / batch_size,
     verbose=1,
     callbacks=[checkpointer, early_stopping, tensorboard]
 )

train/train_street_view.py

@@ -37,7 +37,7 @@
 final_output = Dense(nb_classes, activation='softmax', name='fc360')(x)
 
 # create the new model
-model = Model(input=base_model.input, output=final_output)
+model = Model(inputs=base_model.input, outputs=final_output)
 
 model.summary()
 
@@ -73,8 +73,8 @@
         crop_largest_rect=True,
         shuffle=True
     ),
-    samples_per_epoch=len(train_filenames),
-    nb_epoch=nb_epoch,
+    steps_per_epoch=len(train_filenames) / batch_size,
+    epochs=nb_epoch,
     validation_data=RotNetDataGenerator(
         test_filenames,
         input_shape=input_shape,
@@ -83,7 +83,7 @@
         crop_center=True,
         crop_largest_rect=True
     ),
-    nb_val_samples=len(test_filenames),
+    validation_steps=len(test_filenames) / batch_size,
     callbacks=[checkpointer, early_stopping, tensorboard],
     nb_worker=10,
     pickle_safe=True,

train/train_street_view_regression.py

@@ -35,7 +35,7 @@
 final_output = Dense(1, activation='sigmoid', name='fc1')(x)
 
 # create the new model
-model = Model(input=base_model.input, output=final_output)
+model = Model(inputs=base_model.input, outputs=final_output)
 
 model.summary()
 
@@ -71,8 +71,8 @@
         crop_largest_rect=True,
         shuffle=True
     ),
-    samples_per_epoch=len(train_filenames),
-    nb_epoch=nb_epoch,
+    steps_per_epoch=len(train_filenames) / batch_size,
+    epochs=nb_epoch,
     validation_data=RotNetDataGenerator(
         test_filenames,
         input_shape=input_shape,
@@ -82,7 +82,7 @@
         crop_center=True,
         crop_largest_rect=True
     ),
-    nb_val_samples=len(test_filenames),
+    validation_steps=len(test_filenames) / batch_size,
     callbacks=[checkpointer, early_stopping, tensorboard],
     nb_worker=10,
     pickle_safe=True,