The scripts in this repository can be used to parse the provided annotation files from the Sun Dataset to create import files that will allow you to create CADe and CADx models using Google AutoML. Please choose the script which best fits your use case described below.
-
generate_import_file_object_detection_binary
Create a model which detects polyps within an image and returns a bounding box that frames the polyp.
-
generate_import_file_object_detection_multiclass
Create a model that detects polyps within an image and returns a bounding box that frames the polyp and simultaneously classifies the polyp as belonging to one of the following pathology categories: hyperplastic polyp, sessile serrated lesion, adenoma, serrated adenoma, invasive carcinoma.
Please choose the script below that fits your use case.
-
generate_import_file_binary_polyp_classification
Create a model which classifies an image as either having a polyp or not having a polyp.
-
generate_import_file_multiclass_pathology_polyp_classification
Create a model which classifies an image as not having a polyp or any of the following pathology categories: hyperplastic polyp, sessile serrated lesion, adenoma, serrated adenoma, invasive carcinoma.
Create and download your Google Cloud API credentials to your local machine. Paste the path to the credentials json in the following line of code at the end of the first block in every script.
os.environ['GOOGLE_APPLICATION_CREDENTIALS']=''Additionally, replace the bucket_name variable in the third code block with the name of your Google Cloud Storage Bucket.
storage_client = storage.Client()
bucket_name = 'sun-data-cloud' # replace 'sun-data-cloud' with the name of your Google storage bucket
bucket = storage_client.bucket(bucket_name)