2024Research Project
High-Resolution Urdu OCR
A specialized Optical Character Recognition model built to extract and process complex handwritten Urdu script using C++ data layers and PyTorch engines.
Role
Machine Learning Engineer
Client
Research Project
Year
2024
Duration
4 months
The challenge
Accurately extracting and recognizing unstructured, handwritten Urdu text—a notoriously complex script—poses significant machine learning challenges.
The approach
Engineered and trained a custom OCR model utilizing UTRNet and PyTorch, applying advanced text extraction and robust preprocessing techniques optimized via memory-efficient C++ text pipeline structures.
The outcome
Achieved high accuracy in extracting both printed and handwritten Urdu text formats.
Highlights
- UTRNet Implementation
- PyTorch Model Training
- High Accuracy Extraction
PythonC++PyTorchAI/ML
Next project
Online Billing System