Kuzushiji data collection system for machine learning

Objectives

Manually creating annotation data for machine learning by reading IIIF Manifests as possible as quickly and easily.

Characteristics

• Simple and intuitive user experience.
• Task management with deadline function.
• Automatic generation of bounding boxes on each Kuzushiji characters.
• Automatic annotation on each bounding box if you want.
• Export annotation data in W3C annotation format.

Required tools for this system

• Laravel 9 + Laravel Breeze for user management
• PHP 8.1
• database system (Mysql, Mariadb, PostgreSQL, etc.)
• IIIF manifests.
• For user management, you need to prepare a generic mail function (at least a sending function) on your system.

Special thanks to:

• OpenSeadragon 3.1.0 for IIIF image operation.
• Annotorious 2.7.6 customized to make labels visible directly on Safari + OpenSeadragon plugin as image annotation library + Shape Labels.
• KuroNet, KogumaNet for automatic kuzushiji recognition assistance.

Some Comments

• A journey of a thousand miles begins with a single step.
• Thank you for all developers related to this system.
• It's not only for Kuzushiji. Any IIIF images are applicable.

License

MIT License.

The Laravel framework is open-sourced software licensed under the MIT license.