metadata

{}

Model Card for Model ID

This project aims to create a text scanner that converts paper images into machine-readable formats (e.g., Markdown, JSON). It is the son of Nougat, and thus, grandson of Douat.

The key idea is to combine the bounding box modality with text, achieving a pixel scan behavior that predicts not only the next token but also the next position.

The name "Lougat" is a combination of LLama and Nougat. The key idea is nature continues of this paper [LOCR: Location-Guided Transformer for Optical Character Recognition]([2403.02127] LOCR: Location-Guided Transformer for Optical Character Recognition (arxiv.org))

Current Branch: The LOCR model

Other Branch:

Florence2 + LLama → Flougat
Sam2 + LLama → Slougat
Nougat + Relative Position Embedding LLama → Rlougat

Inference and Train

Please see https://github.com/veya2ztn/Lougat