cvsutils

Unofficial utility scripts for Microsoft Custom Vision Service.

Setup

pip install cvsutils
export CVS_ENDPOINT='<endpoint>'
export CVS_TRAINING_KEY='<training_key>'
export CVS_PREDICTION_KEY='<prediction_key>'
export CVS_PREDICTION_RESOURCE_ID='<resource_id>'

Those keys and endpoint information can be found in the Custom Vision's settings page.

Available commands

# Show a list of projects
cvs_list_projects [--verbose]

# Create a new project
cvs_create_project <dataset_filepath> [--project_name <name>] [--domain_id <domain_id>]

# Download dataset from a project
cvs_download_project <project_id> <output_dir> [--filter_tag <tag_id> [<tag_id> ...]]

# Train a model
cvs_train_project <project_id> [--domain_id <domain_id>] [--type {multiclass,multilabel}] [--force]

# Export a model
cvs_export_model <project_id> <iteration_id> {tensorflow,coreml,onnx} [--output_filepath <filepath>]

And

• cvs_add_images
• cvs_download_predictions
• cvs_evaluate_project
• cvs_get_domains
• cvs_predict_dataset
• cvs_predict_image
• cvs_remove_iteration
• cvs_show_project
• cvs_validate_dataset

To see the detailed help, please run the command with "-h" option.

Dataset file format

This tool uses the SIMPLE dataset format to upload/download datasets from Custom Vision Service.

For details, please see the simpledataset repository.