class AutotrainerTests(unittest.TestCase):
autotrainer: Autotrainer
projects: [Project]
def setUp(self):
self.autotrainer = Autotrainer(CVTK, cv_endpoint, conn_string)
self.autotrainer.blob.initialise_containers()
self.projects = []
def tearDown(self):
for project in self.projects:
print('deleting project id {}'.format(project.id))
self.autotrainer.custom_vision.training_client.delete_project(
project.id)
self.projects.remove(project)
def test_upload_files(self):
image_paths = self.autotrainer.get_file_paths(dog_dir, 'jpg')
self.autotrainer.upload_multiple_images(Container.test, image_paths,
['dog'])
def test_add_images_to_project(self):
project = self.autotrainer.custom_vision.create_project(
'test_add_images', 'created by unit test',
Domain.GENERAL_CLASSIFICATION, ClassificationType.MULTICLASS)
self.projects.append(project) # save to delete later
image_paths = self.autotrainer.get_file_paths(cat_dir, 'jpg')
self.autotrainer.upload_multiple_images(Container.test, image_paths,
['cat'])
self.autotrainer.add_all_images_to_cv(Container.test, project.id)
def test_add_record(self):
test_list = []
test = ImageCreateResult()
test_image = Image()
test_image.id = str(uuid.uuid4())
test.image = test_image
test.source_url = "https://test.com/1.jpg"
test_list.append(test)
test.image.id = str(uuid.uuid4())
test_list.append(test)
self.autotrainer.table.initialise_table(table_name="testAutoTrainer")
self.autotrainer.create_record_of_images(test_list, test_container)
fileName = urlparse(test.source_url).path.split("/")[-1]
record = self.autotrainer.table.get_record(test_container, fileName)
self.assertEqual(record.RowKey, fileName)
self.assertEqual(record.PartitionKey, test_container)
self.autotrainer.table.delete_table()
class AutotrainerTests(unittest.TestCase):
autotrainer: Autotrainer
projects: [Project]
def setUp(self):
self.autotrainer = Autotrainer(CVTK,cv_endpoint,conn_string)
self.autotrainer.blob.initialise_containers()
self.projects = []
def tearDown(self):
for project in self.projects:
print('deleting project id {}'.format(project.id))
self.autotrainer.custom_vision.training_client.delete_project(project.id)
self.projects.remove(project)
def test_upload_files(self):
image_paths = self.autotrainer.get_file_paths(dog_dir, 'jpg')
self.autotrainer.upload_multiple_images(Container.test, image_paths, ['dog'] )
def test_add_images_to_project(self):
project = self.autotrainer.custom_vision.create_project(
'test_add_images',
'created by unit test',
Domain.GENERAL_CLASSIFICATION,
ClassificationType.MULTICLASS )
self.projects.append(project) # save to delete later
image_paths = self.autotrainer.get_file_paths(cat_dir, 'jpg')
self.autotrainer.upload_multiple_images(Container.test, image_paths, ['cat'])
self.autotrainer.add_all_images_to_cv(Container.test, project.id)
def __init__(self):
parser = argparse.ArgumentParser(
description='Autotrainer tools',
usage='autotrainer [cv, catalogue, select] <options>')
parser.add_argument('command', help='Subcommand to run')
# parse_args defaults to [1:] for args, but you need to
# exclude the rest of the args too, or validation will fail
args = parser.parse_args(sys.argv[1:2])
if not hasattr(self, args.command):
print('Unrecognized command')
parser.print_help()
exit(1)
# setup the keys from environment variables
self.cv_key = os.environ['CV_TRAINING_KEY']
self.cv_endpoint = os.environ['CV_ENDPOINT']
self.storage_connection_string = os.environ[
'STORAGE_ACCOUNT_CONNECTION_STRING']
self.autotrainer = Autotrainer(self.cv_key, self.cv_endpoint,
self.storage_connection_string)
getattr(self, args.command)() # call the method on this obj
class AutotrainerCli:
cv_key: str
cv_endpoint: str
storage_connection_string: str
autotrainer: Autotrainer
def __init__(self):
parser = argparse.ArgumentParser(
description='Autotrainer tools',
usage='autotrainer [cv, catalogue, select] <options>')
parser.add_argument('command', help='Subcommand to run')
# parse_args defaults to [1:] for args, but you need to
# exclude the rest of the args too, or validation will fail
args = parser.parse_args(sys.argv[1:2])
if not hasattr(self, args.command):
print('Unrecognized command')
parser.print_help()
exit(1)
# setup the keys from environment variables
self.cv_key = os.environ['CV_TRAINING_KEY']
self.cv_endpoint = os.environ['CV_ENDPOINT']
self.storage_connection_string = os.environ[
'STORAGE_ACCOUNT_CONNECTION_STRING']
self.autotrainer = Autotrainer(self.cv_key, self.cv_endpoint,
self.storage_connection_string)
getattr(self, args.command)() # call the method on this obj
def cv(self):
parser = argparse.ArgumentParser(description='Custom Vision tools',
usage='autotrainer cv <options>')
# prefixing the argument with -- means it's optional
parser.add_argument('--newproject',
help='Name of the new project. Returns project id')
parser.add_argument('--domain',
type=Domain,
choices=list(Domain),
default=Domain.GENERAL_CLASSIFICATION)
parser.add_argument('--type',
type=ClassificationType,
choices=list(ClassificationType),
default=ClassificationType.MULTICLASS)
parser.add_argument('--project',
help='Id of the custom vision project')
parser.add_argument('--train',
help='Train a project. Returns iteration id',
action='store_true')
parser.add_argument('--export',
help='Export a project',
action='store_true')
parser.add_argument('--iteration',
help='(optional) iteration to store')
parser.add_argument('--platform',
help='Platform to export to',
type=Platform,
choices=list(Platform),
default=Platform.DOCKER)
parser.add_argument('--flavour',
help='Platform dependent Flavour',
type=Flavour,
choices=list(Flavour),
default=Flavour.Linux)
parser.add_argument('--listprojects',
help='list custom vision project IDs',
action='store_true')
args = parser.parse_args(sys.argv[2:])
if (args.newproject):
print('Creating new project: ' + args.newproject)
project = self.autotrainer.custom_vision.create_project(
args.newproject, 'Created by autotrainer CLI', args.domain,
args.type)
print(project.id)
elif args.train:
project = self.autotrainer.custom_vision.training_client.get_project(
args.project)
print('Training project: {}'.format(project.name))
iteration = self.autotrainer.custom_vision.train_project_and_wait(
project)
print(iteration.id)
elif args.export:
project = self.autotrainer.custom_vision.training_client.get_project(
args.project)
if (args.iteration):
iteration = self.autotrainer.custom_vision.training_client.get_iteration(
project.id, args.iteration)
else:
iteration = self.autotrainer.custom_vision.training_client.get_iterations(
project.id)[0]
if not iteration.exportable:
print('Iteration is not exportable')
exit(1)
export = self.autotrainer.custom_vision.export_project(
args.platform, args.flavour, project, iteration)
print(export.download_uri)
elif args.listprojects:
ids = self.autotrainer.custom_vision.list_project_ids()
print("{:25s}{:}".format("Project Name", "Project ID"))
print(
"-------------------------------------------------------------"
)
for i in ids:
print("{:25s}{:}".format(i.name, i.id))
else:
print('Incorrect syntax')
def catalogue(self):
# define the CLI args
parser = argparse.ArgumentParser(
description='Data Catalogue tools',
usage='autotrainer catalogue <options>')
parser.add_argument('command',
help='Catalogue options',
type=str,
choices=['describe', 'upload'])
args = parser.parse_args(sys.argv[2:3])
if args.command == 'describe':
parser = argparse.ArgumentParser(
description='Data Catalogue description',
usage='autotrainer catalogue describe')
print('Querying catalogue...')
train_blobs = self.autotrainer.list_all_labelled_blobs(
Container.train, None)
test_blobs = self.autotrainer.list_all_labelled_blobs(
Container.test, None)
holdout_blobs = self.autotrainer.list_all_labelled_blobs(
Container.holdout, None)
print('Training set has {} images'.format(len(train_blobs)))
print_describe_label_frequency(train_blobs)
print('Test set has {} images'.format(len(test_blobs)))
print_describe_label_frequency(test_blobs)
print('Holdout set has {} images'.format(len(holdout_blobs)))
print_describe_label_frequency(holdout_blobs)
elif args.command == 'upload':
parser = argparse.ArgumentParser(
description='Upload to the data catalogue (blob storage)',
usage='autotrainer catalogue upload <options>')
parser.add_argument(
'-d',
'--directory',
help='The local directory containing the images',
required=True)
parser.add_argument('-l',
'--labels',
action='append',
help='Label for the image',
required=True) # can set multiple
parser.add_argument('-c',
'--container',
type=Container,
choices=list(Container),
default=Container.train)
parser.add_argument('--extension',
help='Filter on file extension',
default='')
parser.add_argument('--parent',
help='Parent directory in Blob Storage',
default=None)
args = parser.parse_args(sys.argv[3:])
image_paths = self.autotrainer.get_file_paths(
args.directory, args.extension)
labelled_blobs = self.autotrainer.upload_multiple_images(
args.container, image_paths, args.labels, args.parent)
print('Created {} labelled blobs'.format(len(labelled_blobs)))
def select(self):
# define the CLI args
parser = argparse.ArgumentParser(
description='Select data from blobs and add to Custom Vision',
usage='autotrainer select <options>')
parser.add_argument('-c',
'--container',
type=Container,
choices=list(Container),
default=Container.train)
parser.add_argument('--num',
type=int,
help='Number to add',
required=False)
parser.add_argument('--project',
help='Id of the custom vision project',
required=True)
args = parser.parse_args(sys.argv[2:])
res = self.autotrainer.add_all_images_to_cv(args.container,
args.project, args.num)
res_freq = Counter([i.status for i in res])
print('{} image create results'.format(len(res)))
print('label summary:')
print(res_freq)
def setUp(self):
self.autotrainer = Autotrainer(CVTK,cv_endpoint,conn_string)
self.autotrainer.blob.initialise_containers()
self.projects = []