model = SqueezeNet(nb_classes, input_shape=input_shape)
adam = Adam(lr=0.040)
sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss="categorical_crossentropy",
optimizer='adam',
metrics=['accuracy'])
if os.path.isfile(weights_file):
print('Loading weights: %s' % weights_file)
model.load_weights(weights_file, by_name=True)
print('Fitting model')
# model.fit(images, classes, batch_size=batch_size, nb_epoch=nb_epoch, verbose=1, validation_split=0.2, initial_epoch=0)
model.fit_generator(training_data,
samples_per_epoch=samples_per_epoch,
validation_data=validation_data,
nb_val_samples=nb_val_samples,
nb_epoch=nb_epoch,
verbose=1,
initial_epoch=initial_epoch)
print("Finished fitting model")
print('Saving weights')
model.save_weights(weights_file, overwrite=True)
print('Evaluating model')
# score = model.evaluate(images, classes, verbose=1)
# validation_data = ( (load_image(x), to_categorical([y], nb_classes=nb_classes)) for x, y in test )
# validation_data = gen(X_test, Y_test)
score = model.evaluate_generator(validation_data, val_samples=nb_val_samples)
# score = model.evaluate(X_test, Y_test, verbose=1)
print('result: %s' % score)
# log progress to AML workspace
if remote_execution:
class LogRunMetrics(Callback):
# callback at the end of every epoch
def on_epoch_end(self, epoch, log):
# log a value repeated which creates a list
run.log('val_loss', log['val_loss'])
run.log('loss', log['loss'])
callbacks.append(LogRunMetrics())
model.fit_generator(train_generator,
steps_per_epoch=50,
epochs=30,
verbose=1,
callbacks=callbacks,
validation_data=val_generator,
validation_steps=80,
workers=4)
if remote_execution:
run.log('final_val_loss', model.history.history['val_loss'][-1])
run.log('final_val_accuracy', model.history.history['val_accuracy'][-1])
# # Loss/epoch plots
# # Loss/epoch plots
# In[19]:
plt.plot(model.history.history['categorical_crossentropy'], label='train')
plt.plot(model.history.history['val_categorical_crossentropy'], label='val')
print('Loading weights: %s' % weights_file)
model.load_weights(weights_file, by_name=True)
else:
# if it is a new model, set balanced bias for the last layer
w, _ = model.get_layer('conv10').get_weights()
model.get_layer('conv10').set_weights(
[w, np.log(class_weight / sum(class_weight))])
print('Fitting model')
# Balance class weights by frequency and give preference for RED lights.
class_weight = class_weight.compute_class_weight(
'balanced', np.unique(train_generator.classes), train_generator.classes)
class_weight[0] *= 2
model.fit_generator(train_generator,
samples_per_epoch=samples_per_epoch,
validation_data=val_generator,
validation_steps=nb_val_samples // 64,
nb_epoch=nb_epoch,
verbose=1,
initial_epoch=initial_epoch,
class_weight=class_weight)
print("Finished fitting model")
print('Saving weights')
model.save_weights(weights_file, overwrite=True)
print('Evaluating model')
score = model.evaluate_generator(val_generator,
steps=int(samples_per_epoch / nb_val_samples))
print('result: %s' % score)
# Convert class vectors to binary class matrices.
y_train = keras.utils.to_categorical(y_train, num_classes)
x_train = x_train.astype(np.float32)
x_train /= 255.
img_shape = x_train[0].shape
model = SqueezeNet(img_shape, num_classes)
# This will do preprocessing and realtime data augmentation:
datagen = ImageDataGenerator(
featurewise_center=False, # set input mean to 0 over the dataset
samplewise_center=False, # set each sample mean to 0
featurewise_std_normalization=False, # divide inputs by std of the dataset
samplewise_std_normalization=False, # divide each input by its std
zca_whitening=False, # apply ZCA whitening
rotation_range=0, # randomly rotate images in the range (degrees, 0 to 180)
width_shift_range=
0.1, # randomly shift images horizontally (fraction of total width)
height_shift_range=
0.1, # randomly shift images vertically (fraction of total height)
horizontal_flip=True, # randomly flip images
vertical_flip=False) # randomly flip images
datagen.fit(x_train)
#model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, shuffle=True)
model.fit_generator(datagen.flow(x_train, y_train, batch_size=batch_size),
steps_per_epoch=x_train.shape[0] // batch_size,
epochs=epochs)
model.save(os.path.join(path, 'model.h5'))
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--train-dir', default='data/train')
parser.add_argument('--test-dir', default='data/test')
parser.add_argument('--logdir', default='logs')
parser.add_argument('--batch-size', type=int, default=32)
parser.add_argument('--epochs', type=int, required=True)
parser.add_argument('--num-classes', type=int, required=True)
parser.add_argument('--checkpoint-pattern',
default='weights-{epoch:d}-{val_acc:.4f}.hdf5')
parser.add_argument('--learning-rate', type=float, default=1e-4)
parser.add_argument('--restore')
args = parser.parse_args()
# count samples
train_files = count_files(args.train_dir, '.png')
print('Found %d train files.' % train_files)
test_files = count_files(args.test_dir, '.png')
print('Found %d test files.' % test_files)
if args.restore:
model = SqueezeNet(weights=None, classes=args.num_classes)
model.load_weights(args.restore)
else:
model = SqueezeNet(weights='imagenet', classes=args.num_classes)
model.compile(loss='categorical_crossentropy',
optimizer=optimizers.Adam(lr=args.learning_rate),
metrics=['accuracy'])
train_datagen = ImageDataGenerator(
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,
preprocessing_function=preprocess_single)
test_datagen = ImageDataGenerator(preprocessing_function=preprocess_single)
train_generator = train_datagen.flow_from_directory(
args.train_dir,
target_size=(SIZE, SIZE),
batch_size=args.batch_size,
class_mode='categorical')
test_generator = test_datagen.flow_from_directory(
args.test_dir,
target_size=(SIZE, SIZE),
batch_size=args.batch_size,
class_mode='categorical')
checkpoint = ModelCheckpoint(args.checkpoint_pattern,
monitor='val_acc',
verbose=1,
save_best_only=True,
mode='max')
tensorboard = TensorBoard(log_dir=args.logdir,
histogram_freq=0,
batch_size=args.batch_size,
write_graph=True,
write_grads=False,
write_images=False,
embeddings_freq=0,
embeddings_layer_names=None,
embeddings_metadata=None)
callbacks = [checkpoint, tensorboard]
model.fit_generator(train_generator,
steps_per_epoch=(train_files // args.batch_size),
epochs=args.epochs,
validation_data=test_generator,
validation_steps=(test_files // args.batch_size),
callbacks=callbacks)