Logo detection example running on a small set of custom-collected data. Grabs YOLO-annotations and images, repackages them into SFrames, runs and validates Turi Create training job on SageMaker and pushes the validated .mlmodel to an REST API that handles file downloads.

In addition, a dedicated pipeline step generates a model card for the purposes for AIGA WP3. A publicly available model card automatically generated from the last model execution is available on this static page.

Currently, the code assumes that all non-SageMaker resources - Lambda functions, S3 buckets and training data in them - are already provisioned on AWS.

The pipeline can be updated and run as follows, from under the pipelines folder (TODO: add a setup.py file):

from pipeline import pipeline, conf

# updates the pipeline definition on SageMaker
pipeline.upsert(role_arn=conf.role)

# starts a new pipeline execution
pipeline.start()

To adapt to your own needs, the step functions - processing and training need to be replaced with the ones that best suit your model training. So does the config.ini file in the pipelines folder, which is a convenience shortcut for configuration variables.