Features Pipeline Models Get Started Profile GitHub Docs

Build Your Own
OCR Pipeline

Open-source document understanding framework. Complete training scripts, modular architecture, and full control. Train your own models.

🚀 Open Source 📝 End-to-End OCR ⚡ 100+ Models 🏗️ Modular 💪 DDP Support

Why Doctane?

Complete control over your document processing pipeline

📝

End-to-End OCR

Complete pipeline from text detection to recognition. Handles straight and rotated text seamlessly.

🔄

Layout Understanding

Automatic page orientation detection and straightening. Multi-column document support.

🏗️

Modular Architecture

Plug-and-play model components. Easy to extend with custom models and pipelines.

⚡

High Performance

Optimized inference with PyTorch. DDP training support for multi-GPU acceleration.

🌐

Multi-Language

Support for English, French, and extensible to other languages.

📊

Structured Output

Hierarchical Document objects. Export to JSON and hOCR formats.

Processing Pipeline

Four-stage pipeline for complete document understanding

1

Preprocessing

Page orientation detection and optional straightening.

2

Text Detection

Segmentation-based model identifies text regions.

3

Text Recognition

Crops fed to recognition model for transcription.

4

Document Assembly

Structured output with geometry and confidence scores.

100+ Supported Models

State-of-the-art architectures for detection and recognition

LinkNet

Detection

DeepLabV3+

Detection

SegFormer

Detection

UNet

Detection

UNet++

Detection

FPN

Detection

PSPNet

Detection

PAN

Detection

MAnet

Detection

Faster R-CNN

Detection

SAR

Recognition

ViTSTR

Recognition

CRNN

Recognition

MASTER

Recognition

TRBA

Recognition

ABINet

Recognition

LSTR

Recognition

ViTPTR

Recognition

MATRN

Recognition

PARSeq

Recognition

Plus 80+ encoder variants (ResNet, EfficientNet, VGG, MobileNet, DenseNet, etc.)

Get Started

Clone, install, and run in minutes

# Clone the repository git clone https://github.com/Purushothaman-natarajan/doctane.git cd doctane # Install dependencies pip install -r requirements.txt # Start the API server python api/main.py
1

Clone

Download from GitHub

2

Install

Run pip install

3

Launch

Open localhost:8000/app

🔧 No Pre-trained Weights

We provide the code only, not model weights. Train your own models using the provided scripts.

💻 Bring Your Infrastructure

Training requires GPU. Use your own hardware or cloud (AWS/GCP/Azure). DDP supported.

👨‍💻

Purushothaman Natarajan

AI Engineer & Full-Stack Developer specializing in Computer Vision, NLP, and Deep Learning systems.

GitHub Portfolio Email

Featured Projects

🔬 Doctane

Multimodal intelligent document analysis and understanding system with OCR, layout understanding.

Python PyTorch OCR

🔒 Exploit2Patch

AI-Powered Vulnerability Intelligence Platform with autonomous CVE research and patch generation.

AI Agents Cybersecurity

🧪 DL-Studio

Local deep learning development environment with 20+ algorithms, built-in XAI, and web interface.

Deep Learning XAI Streamlit