def run_acdml():
acdml = AttentionCrossDomainMetaLearning(
database=None,
val_database=ISICDatabase(),
test_database=EuroSatDatabase(),
network_cls=get_assembled_model,
n=5,
k_ml=1,
k_val_ml=5,
k_val=1,
k_val_val=15,
k_test=1,
k_val_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,
report_validation_frequency=1000,
log_train_images_after_iteration=1000,
num_tasks_val=100,
clip_gradients=True,
experiment_name='acdml_11',
val_seed=42,
val_test_batch_norm_momentum=0.0,
)
acdml.train(iterations=60000)
acdml.evaluate(iterations=100, num_tasks=1000, seed=14)
def run_separate_domain():
test_database = ISICDatabase()
maml = SeparateDomainMAML(
database=test_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=15000,
meta_learning_rate=0.001,
report_validation_frequency=1000,
log_train_images_after_iteration=1000,
number_of_tasks_val=100,
number_of_tasks_test=1000,
clip_gradients=True,
experiment_name='separate_domain_isic',
val_seed=42,
val_test_batch_norm_momentum=0.0,
)
maml.train(iterations=60000)
maml.evaluate(100, seed=14)
def run_transfer_learning():
isic_database = ISICDatabase()
transfer_learning = TransferLearningVGG16(
database=isic_database,
n=5,
k_val_test=15,
k_test=5,
lr_inner_ml=0.01,
number_of_tasks_test=100,
val_test_batch_norm_momentum=0.0,
random_layer_initialization_seed=42,
num_trainable_layers=0,
)
transfer_learning.evaluate(10, seed=42, use_val_batch_statistics=True)
def get_test_dataset(self, seed=-1):
database = ISICDatabase()
test_dataset = self.get_supervised_meta_learning_dataset(
database.test_folders,
n=self.n,
k=self.k_test,
k_validation=self.k_val_test,
meta_batch_size=1,
seed=seed
)
test_dataset = test_dataset.repeat(-1)
test_dataset = test_dataset.take(self.number_of_tasks_test)
setattr(test_dataset, 'steps_per_epoch', self.number_of_tasks_test)
return test_dataset
def get_val_dataset(self):
database = ISICDatabase()
val_dataset = self.get_supervised_meta_learning_dataset(
database.val_folders,
n=self.n,
k=self.k,
k_validation=self.k_val_val,
meta_batch_size=1,
seed=self.val_seed,
)
val_dataset = val_dataset.repeat(-1)
val_dataset = val_dataset.take(self.number_of_tasks_val)
setattr(val_dataset, 'steps_per_epoch', self.number_of_tasks_val)
return val_dataset
def get_val_dataset(self):
databases = [ISICDatabase()]
val_dataset = self.get_cross_domain_meta_learning_dataset(
databases=databases,
n=self.n,
k=self.k,
k_validation=self.k_val_val,
meta_batch_size=1,
partition='val')
val_dataset = val_dataset.repeat(-1)
val_dataset = val_dataset.take(self.number_of_tasks_val)
setattr(val_dataset, 'steps_per_epoch', self.number_of_tasks_val)
return val_dataset
def setUp(self):
def parse_function(example_address):
return example_address
self.parse_function = parse_function
self.omniglot_database = OmniglotDatabase(random_seed=-1,
num_train_classes=1200,
num_val_classes=100)
self.mini_imagenet_database = MiniImagenetDatabase()
self.celeba_database = CelebADatabase()
self.lfw_database = LFWDatabase()
self.euro_sat_database = EuroSatDatabase()
self.isic_database = ISICDatabase()
self.chest_x_ray_database = ChestXRay8Database()
if __name__ == '__main__':
tf.config.set_visible_devices([], 'GPU')
root_folder_to_save = os.path.expanduser('~/datasets_visualization/')
if not os.path.exists(root_folder_to_save):
os.mkdir(root_folder_to_save)
databases = (
MiniImagenetDatabase(),
OmniglotDatabase(random_seed=42,
num_train_classes=1200,
num_val_classes=100),
AirplaneDatabase(),
CUBDatabase(),
DTDDatabase(),
FungiDatabase(),
VGGFlowerDatabase(),
# TrafficSignDatabase(),
MSCOCODatabase(),
PlantDiseaseDatabase(),
EuroSatDatabase(),
ISICDatabase(),
ChestXRay8Database(),
)
# for database in databases:
# visualize_database(database, root_folder_to_save)
visualize_all_domains_together(databases, root_folder_to_save)
]
model = tf.keras.models.Model(inputs=vgg19.inputs, outputs=outputs)
def convert_to_activaitons(imgs):
imgs_activations = []
for image in imgs:
image = convert_str_to_img(image)[np.newaxis, ...]
activations = model.predict(image)
imgs_activations.append(activations)
return imgs_activations
DTDDatabase()
d1_imgs = get_all_instances(MiniImagenetDatabase())
d2_imgs = get_all_instances(ISICDatabase())
d3_imgs = get_all_instances(ChestXRay8Database())
d4_imgs = get_all_instances(PlantDiseaseDatabase())
d1_imgs = np.random.choice(d1_imgs, 10, replace=False)
d2_imgs = np.random.choice(d2_imgs, 10, replace=False)
d3_imgs = np.random.choice(d3_imgs, 10, replace=False)
d4_imgs = np.random.choice(d4_imgs, 10, replace=False)
print(d1_imgs)
print(d2_imgs)
print(d3_imgs)
print(d4_imgs)
d1_imgs_activations = convert_to_activaitons(d1_imgs)
d2_imgs_activations = convert_to_activaitons(d2_imgs)