MyOCR is an advanced OCR pipeline builder designed for engineers and researchers to create and integrate custom OCR systems. It offers a modular and extensible framework for end-to-end OCR development, enabling flexible training, integration of deep learning models, and production-ready deployment.

How It Works

MyOCR provides a unified pipeline for detection and recognition, allowing users to mix and match components like models and processors. It leverages ONNX runtime for efficient CPU/GPU inference and supports structured OCR output through integration with large language models like Qwen for data extraction.

Quick Start & Requirements

• Install: pip install -e . (after cloning the repo)
• Requirements: Python 3.11+, CUDA 12.6+ recommended for GPU.
• Setup: Clone repo, install dependencies, download pre-trained weights.
• Docs: https://robbyzhaox.github.io/myocr/
• Demo: https://huggingface.co/spaces/robbyzhaox/myocr

Highlighted Details

• End-to-end OCR development framework with modular components.
• Developer-friendly Python APIs and prebuilt pipelines.
• ONNX runtime support for fast CPU/GPU inference.
• Structured OCR output via LLM integration (Ollama, OpenAI).

Maintenance & Community

• Active development with recent releases (v0.1.1 on May 17, 2025).
• Contribution guidelines provided.

Licensing & Compatibility

• Licensed under Apache 2.0.
• Permissive license suitable for commercial use and closed-source integration.

Limitations & Caveats

The structured output pipeline requires configuration for LLM APIs (Ollama, OpenAI) and specific model setups. The README mentions a UI (doc-insight-ui) but does not provide a direct link.