trocr

Large-scale Self-supervised Pre-training Across Tasks, Languages, and Modalities

Powerful handwritten text recognition. A simple-to-use, unofficial implementation of the paper "TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models".

Handwritten mathematical symbols recognition with TrOCR

本项目是一个结合了深度学习与自动化流程的工具，用于处理极验（Geetest）v4 的文字点选验证码。它使用 **YOLOv8** 模型进行目标检测，并集成了 TrOCR 和 PaddleOCR 模型进行文字识别，提供了灵活的配置和强大的混合匹配策略。

Modular OCR pipeline for live-camera documents/screens. Auto-aligns perspective, blocks obstructions, and extracts text using interchangeable deep-learning models.

myNRC-OCR: A Hybrid OCR Framework for Handwritten Myanmar NRCs Integrating YOLOv8, Tesseract LSTM, TrOCR, and Post-OCR Correction

AutoRegressive Transformer Model for Optical Character Recognition In Farasi

An image to text model base on transformer which can also be used on OCR task.

FrameReader is a full-fledged service for recognizing text on frames of video materials in Russian, training, deployment of computer vision models to solve the OCR problem. Optimization of models on the tensorrt engine

OCR-D processor for TrOCR

A safer way to organize your handwritten medical prescriptions with the functionality of blockchain.

A service for extracting and indexing archival document images

In this project, I used Yolo V8 by Ultralytics to make this project work.The application’s task is to extract car plate numbers accurately by using the techniques mentioned above.

A handwriting recognition app built around the pretrained microsoft/trocr-base-handwritten model and "mltu" CRNN model. The project focuses on model comparison for education and future improvement.

Doxaria OCR/HTR for medical insurance documents

AI-powered automated grading system using TrOCR, Sentence-BERT, OpenCV, and Streamlit to evaluate handwritten answer sheets through semantic analysis, keyword matching, and diagram comparison.

An AI-powered Optical Character Recognition (OCR) platform specifically optimized for multi-line cursive handwriting using TrOCR, vertical center-density line segmentation, and spelling correction.

High-performance LPR system optimized for Indian license plates, achieving 97% character accuracy. Features a hybrid pipeline using YOLOv11 and Mamba-SSM (State Space Models) with built-in regex correction and Beam Search decoding.

This is a handwritten-equations to latex-equations converter, built using trocrbasestage1 model + CROHME dataset + custom-built SentencePiece Tokenizer (latex)

MNIST Sequence Image-to-text TrOCR transformer using Hugging Face