This model classifies text recognized from images into distinct categories based on textual content and features. It combines BERT embeddings with numeric features to make predictions, leveraging both textual and contextual information.
Classification Labels
The model assigns text to the following categories:
HEADER
FOOTER
ANNOTATION
TITLE
CHAPTER_TITLE
CHAPTER_TEXT
OTHER
ANNOTATION_REFERENCE
EPIGRAPH
AUTHOR
Key Features
Hybrid Model: Combines BERT for text embeddings with additional numeric features (e.g., bounding boxes, font size, etc.).
Customizable Architecture: Easy to adapt to other classification tasks with similar structure.
Efficient Training Pipeline: Handles JSON datasets with labeled text for training.
