def create_model():
# Set up standard WideResNet-16-10 model.
model = WideResidualNetwork(depth=16,
width=10,
weights=None,
input_shape=input_shape,
classes=num_classes,
dropout_rate=0.01)
# WideResNet model that is included with Keras is optimized for inference.
# Add L2 weight decay & adjust BN settings.
model_config = model.get_config()
for layer, layer_config in zip(model.layers, model_config['layers']):
if hasattr(layer, 'kernel_regularizer'):
regularizer = keras.regularizers.l2(args.wd)
layer_config['config']['kernel_regularizer'] = \
{'class_name': regularizer.__class__.__name__,
'config': regularizer.get_config()}
if type(layer) == keras.layers.BatchNormalization:
layer_config['config']['momentum'] = 0.9
layer_config['config']['epsilon'] = 1e-5
model = keras.models.Model.from_config(model_config)
opt = keras.optimizers.SGD(lr=args.base_lr)
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=opt,
metrics=['accuracy'])
return model
def resnet(input_shape, num_classes, wide=False):
if wide:
model = WideResidualNetwork(depth=28,
width=8,
dropout_rate=0.0,
weights=None,
input_shape=input_shape,
classes=num_classes)
else:
model = ResNet18(input_shape=input_shape, classes=num_classes)
model.summary()
return model
featurewise_std_normalization=True)
test_gen.mean = train_gen.mean
test_gen.std = train_gen.std
test_iter = test_gen.flow(x_test, y_test, batch_size=args.val_batch_size)
# Restore from a previous checkpoint, if initial_epoch is specified.
# Horovod: restore on the first worker which will broadcast both model and optimizer weights
# to other workers.
if resume_from_epoch > 0 and hvd.rank() == 0:
model = hvd.load_model(
args.checkpoint_format.format(epoch=resume_from_epoch))
else:
# Set up standard WideResNet-16-10 model.
model = WideResidualNetwork(depth=16,
width=10,
weights=None,
input_shape=input_shape,
classes=num_classes,
dropout_rate=0.01)
# WideResNet model that is included with Keras is optimized for inference.
# Add L2 weight decay & adjust BN settings.
model_config = model.get_config()
for layer, layer_config in zip(model.layers, model_config['layers']):
if hasattr(layer, 'kernel_regularizer'):
regularizer = keras.regularizers.l2(args.wd)
layer_config['config']['kernel_regularizer'] = \
{'class_name': regularizer.__class__.__name__,
'config': regularizer.get_config()}
if type(layer) == keras.layers.BatchNormalization:
layer_config['config']['momentum'] = 0.9
layer_config['config']['epsilon'] = 1e-5
tempY = testY
trainY = kutils.to_categorical(trainY)
testY = kutils.to_categorical(testY)
generator = ImageDataGenerator(
rotation_range=10,
width_shift_range=5.0 / 32,
height_shift_range=5.0 / 32,
horizontal_flip=True,
)
generator.fit(trainX, seed=0, augment=True)
# We will be training the model, therefore no need to load weights
model = WideResidualNetwork(depth=28, width=8, dropout_rate=0.0, weights=None)
model.summary()
model.compile(loss="categorical_crossentropy",
optimizer="adam",
metrics=["acc"])
print("Finished compiling")
model_checkpoint = callbacks.ModelCheckpoint(
"WRN-28-8 Weights.h5",
monitor="val_acc",
save_best_only=True,
save_weights_only=True,
)
model.fit_generator(
generator.flow(trainX, trainY, batch_size=batch_size),
testX = testX.astype('float32')
testX /= 255.0
tempY = testY
trainY = kutils.to_categorical(trainY)
testY = kutils.to_categorical(testY)
generator = ImageDataGenerator(rotation_range=10,
width_shift_range=5. / 32,
height_shift_range=5. / 32,
horizontal_flip=True)
generator.fit(trainX, seed=0, augment=True)
# We will be training the model, therefore no need to load weights
model = WideResidualNetwork(depth=28, width=8, dropout_rate=0.0, weights=None)
model.summary()
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['acc'])
print('Finished compiling')
model.fit_generator(generator.flow(trainX, trainY, batch_size=batch_size),
steps_per_epoch=len(trainX) // batch_size,
epochs=epochs,
callbacks=[
callbacks.ModelCheckpoint('WRN-28-8 Weights.h5',
monitor='val_acc',
save_best_only=True,
testX = testX.astype('float32')
testX /= 255.0
tempY = testY
trainY = kutils.to_categorical(trainY)
testY = kutils.to_categorical(testY)
generator = ImageDataGenerator(rotation_range=10,
width_shift_range=5. / 32,
height_shift_range=5. / 32,
horizontal_flip=True)
generator.fit(trainX, seed=0, augment=True)
# We will be training the model, therefore no need to load weights
model = WideResidualNetwork(depth=28, width=8, dropout_rate=0.0, weights=None)
model.summary()
model.compile(loss="categorical_crossentropy",
optimizer="adam",
metrics=["accuracy"])
print("Finished compiling")
model.fit_generator(
generator.flow(trainX, trainY, batch_size=batch_size),
samples_per_epoch=len(trainX),
nb_epoch=nb_epoch,
callbacks=[
callbacks.ModelCheckpoint("WRN-28-8 Weights.h5",
monitor="val_acc",