About GeoDraw
GeoDraw facilitates the handwriting practice of letters by using Image Recognition Technology.
You can assess letter recognition on the unaffiliated site: GrowingReadersandWriters.com
Handwriting
- Produces varied renditions of letters
- Observing varied renditions of letters improves letter recognition
- Promotes learning by activating different parts of the brain to work together
- Sensori-motor Feedback
- Tactile Feedback
- Visual-Temporal Feedback
Tracing and Typing
- Do not produce varied rendions of letters
- Do not activate as many parts of the brain
Technical Information
The Letter and Digit classifiers were trained using the EMNIST Dataset. The EMNIST Dataset consists of over 800,000 handwriteen characters and digits.
The Shape classifier was trained using a dataset of 2,000 computer-drawn shapes.
Project Architecture
GeoDraw was developed using the Flask web framework. Hosted on a GCP Virtual Instance, Docker Compose is used to organize three Docker containers: Nginx, uWSGI serving Flask, and TensorFlow Serving.
Advantages of TensorFlow Serving
Deploying the trained models using TensorFlow Serving has many benefits over serving with TensorFlow.js or TensorFlow Lite. TensorFlow Serving provides a centralized and scalable serving platform accessed by web requests. Multiple models (and model versions) can be served simultaeously. All clients have the same inference experience because model updates are global and immediate.