def run_cub():
cub_database = CUBDatabase()
base_model = tf.keras.applications.VGG19(weights='imagenet')
feature_model = tf.keras.models.Model(inputs=base_model.input,
outputs=base_model.layers[24].output)
sml = SML(
database=cub_database,
network_cls=MiniImagenetModel,
n=5,
k=1,
k_val_ml=5,
k_val_val=15,
k_val_test=15,
k_test=15,
meta_batch_size=4,
num_steps_ml=5,
lr_inner_ml=0.05,
num_steps_validation=5,
save_after_iterations=3000,
meta_learning_rate=0.001,
n_clusters=5000,
feature_model=feature_model,
# feature_size=288,
feature_size=4096,
input_shape=(224, 224, 3),
preprocess_function=tf.keras.applications.vgg19.preprocess_input,
log_train_images_after_iteration=1000,
number_of_tasks_val=100,
number_of_tasks_test=1000,
clip_gradients=True,
report_validation_frequency=250,
experiment_name='cub_imagenet_features')
# sml.train(iterations=6000)
sml.evaluate(iterations=50, seed=42, iterations_to_load_from=3000)
def run_celeba():
vox_celeb_database = VoxCelebDatabase()
feature_model = hub.Module("https://tfhub.dev/google/speech_embedding/1")
sml = SML(
database=vox_celeb_database,
network_cls=VoxCelebModel,
n=5,
k_ml=1,
k_val_ml=5,
k_val_val=15,
k_val_test=15,
k_test=15,
meta_batch_size=4,
num_steps_ml=5,
lr_inner_ml=0.05,
num_steps_validation=5,
save_after_iterations=15000,
meta_learning_rate=0.001,
n_clusters=20,
feature_model=feature_model,
# feature_size=288,
feature_size=4096,
input_shape=(224, 224, 3),
preprocess_function=tf.keras.applications.vgg19.preprocess_input,
log_train_images_after_iteration=1000,
number_of_tasks_val=100,
number_of_tasks_test=1000,
clip_gradients=True,
report_validation_frequency=250,
experiment_name='voxceleb_embedding_features')
sml.train(iterations=60000)
sml.evaluate(iterations=50, seed=42)
def run_cub():
cub_database = CUBDatabase()
base_model = tf.keras.applications.InceptionResNetV2(weights=None)
feature_model = tf.keras.models.Model(inputs=base_model.input,
outputs=base_model.layers[-2].output)
sml = SML(
database=cub_database,
network_cls=MiniImagenetModel,
n=5,
k=1,
k_val_ml=5,
k_val_val=15,
k_val_test=15,
k_test=1,
meta_batch_size=4,
num_steps_ml=5,
lr_inner_ml=0.05,
num_steps_validation=5,
save_after_iterations=2000,
meta_learning_rate=0.001,
n_clusters=500,
feature_model=feature_model,
# feature_size=288,
# feature_size=4096,
feature_size=1536,
input_shape=(224, 224, 3),
preprocess_function=tf.keras.applications.inception_resnet_v2.
preprocess_input,
log_train_images_after_iteration=1000,
number_of_tasks_val=100,
number_of_tasks_test=1000,
clip_gradients=True,
report_validation_frequency=250,
experiment_name='cub_resnet_features_unsupervised')
sml.train(iterations=20000)
sml.evaluate(iterations=50, seed=42)
def run_omniglot():
omniglot_database = OmniglotDatabase(random_seed=47,
num_train_classes=1200,
num_val_classes=100)
base_model = tf.keras.applications.VGG19(weights='imagenet')
feature_model = tf.keras.models.Model(inputs=base_model.input,
outputs=base_model.layers[24].output)
sml = SML(
database=omniglot_database,
network_cls=SimpleModel,
n=5,
k=1,
k_val_ml=5,
k_val_val=15,
k_val_test=15,
k_test=1,
meta_batch_size=4,
num_steps_ml=5,
lr_inner_ml=0.05,
num_steps_validation=5,
save_after_iterations=15000,
meta_learning_rate=0.001,
n_clusters=5000,
feature_model=feature_model,
# feature_size=288,
feature_size=4096,
input_shape=(224, 224, 3),
preprocess_function=tf.keras.applications.vgg19.preprocess_input,
log_train_images_after_iteration=1000,
number_of_tasks_val=100,
number_of_tasks_test=1000,
clip_gradients=True,
report_validation_frequency=250,
experiment_name='omniglot_imagenet_features')
sml.train(iterations=60000)
sml.evaluate(iterations=50, seed=42)