VitaminaOCR · Data, detection and screen reading
Tools for annotation, data preparation, training and OCR evaluation, evolving from mobile automation towards interface reading for BookedUp.
Project Description
VitaminaOCR builds on earlier mobile-automation and screen-reading work, with OCR code located in April 2026. This entry describes the BookedUp-oriented version located in September 2026.
The earlier project detected regions, recognised text and located screen elements to coordinate mobile interaction. I extended that work into tools for importing, annotating and reviewing images, preparing datasets, training and evaluation.
The implementation uses FastAPI, YOLO and OCR components to compare detected regions and recognised text. Dataset splitting keeps identical images together to avoid duplicate examples leaking across evaluation sets. The project continues evolving for BookedUp interface automation.
Main Features
- Evolution from mobile automation
- Image annotation and review
- Dataset preparation and versioning
- Detection and OCR training
- Region and text evaluation
- BookedUp interface reading