An autotraining tool for customvision.ai using Azure Blob Storage and Azure Pipelines

• Data Catalogue
• Train, Test, and Holdout sets
• CLI, Python and CI/CD friendly
• Get started quickly with local uploads
• Store labels with images
• Select data for training Custom Vision models
• Train and export your models

Backed by Azure Storage, Autotrainer helps you maintain a large collection of labelled images for machine learning.

Machine learning often requires the use of multiple datasets that must remain segregated. Autotrainer provides three containers for image datasets: train, test, and holdout.

• Train: Used to training the model.
• Test: Used to test the model during training, and potentially join the training set.
• Holdout: Validate your model using this unseen data.

Consume autotrainer via the CLI, in Python code, or run in Azure Pipelines.

Upload a set of images from a directory in a single command.

Labels are stored in special label files, right next to the image in blob storage.

Select images from your training set, and push them to a Custom Vision project.

Automate the training and exporting of models.

docker run -it -e "CV_ENDPOINT=https://southcentralus.api.cognitive.microsoft.com" -e "CV_TRAINING_KEY=your_key" -e "STORAGE_ACCOUNT_CONNECTION_STRING=your_connection_string" flanagan89/custom-vision-autotrainer -h

Autotrainer requires three environment variables:

• CV_ENDPOINT : The location of your custom vision service, e.g. https://southcentralus.api.cognitive.microsoft.com
• CV_TRAINING_KEY : Your custom vision training key
• STORAGE_ACCOUNT_CONNECTION_STRING : Connection string to an Azure Storage Account

I recommend using Mini Conda to manage your python environment. Download and install miniconda, then in a shell:

1. Create a conda environment: conda create -n customvisionautotrainer python=3.6
2. Activate the environment: activate customvisionautotrainer.
3. Install runtime dependencies: pip install -r src/autotrainer/requirements.txt
4. Install developer dependencies: pip install -r src/autotrainer/requirements-dev.txt
5. Configure environment variables (see above)

$ cd src/autotrainer
$ python ./autotrainer_cli.py -h
usage: autotrainer [cv, catalogue, select] <options>

Autotrainer tools

positional arguments:
  command     Subcommand to run

optional arguments:
  -h, --help  show this help message and exit

$ python ./autotrainer_cli.py catalogue -h
usage: autotrainer catalogue <options>

Data Catalogue tools

positional arguments:
  {describe,upload}  Catalogue options

optional arguments:
  -h, --help         show this help message and exit

Autotrainer uses nose

First, run Azurite for local blob storage testing:

$ cd src/
$ docker-compose up -d

Then you can run the tests.

$ cd src/autotrainer
$ nosetests
............
----------------------------------------------------------------------
Ran 12 tests in 7.088s

OK

NOTE: Some tests require access to a real Azure Storage account and Custom Vision service. See the environment variable section above.

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.microsoft.com.

When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact [email protected] with any additional questions or comments.