docTR

docTR is a free Python library for detecting and recognizing text in documents, with layout analysis and tools for custom model training. It suits developers and researchers building document-automation workflows who can work in Python; its main trade-off is a Python 3.11-or-higher requirement.

Overview

Rather than a ready-made desktop OCR app, docTR is a library for adding OCR to software and research pipelines. Its pretrained predictors locate text and recognize it in a two-stage process, while a separate layout predictor identifies regions such as tables, figures, and headers. That combination makes it useful when a workflow needs more than extracted text, though integrating a library is a different fit from using a standalone OCR service.

Key features

• Document OCR: Pretrained detection and recognition predictors process PDFs and images. Results can be rendered as plain text or exported as a nested, JSON-serializable dictionary, giving developers a practical route from source documents to application data.
• Layout analysis: A predictor identifies document regions including tables, figures, and headers. This is valuable when downstream processing needs to distinguish page structure, not just read words.
• Flexible inputs: The quickstart covers PDFs, images, and web pages; HTML input requires the optional html extra. That breadth helps with mixed-source workflows, but HTML users must account for the extra dependency.
• Model customization: Users can train detection, recognition, layout, and table-structure models when pretrained ones fall short. This provides room to adapt the library to specialized documents, at the cost of taking on model-training work.
• Deployment and integration: Models can be loaded and shared through the Hugging Face Hub. A minimal REST API template and browser demo offer starting points for serving or demonstrating OCR, rather than a turnkey hosted service.
• Additional tools: The contrib module includes an artefact detector for logos, QR codes, and barcodes, but it requires onnxruntime. Pretrained weights download on first use and are cached locally for subsequent calls.

The project describes its predictors as optimized for CPU and GPU inference. Installation is through pip, Git, or official Docker images; the Python 3.11-or-higher requirement may rule it out for environments that cannot meet that baseline. For AWS Lambda, the guide calls for disabling multiprocessing and placing the model cache in /tmp to meet the platform's write restrictions.

Pricing

docTR is free, with no free trial because there is no paid plan to trial. The open-source Python library costs nothing and includes the library for Python 3.11 or higher, installable with pip, Git, or Docker. Its Apache License 2.0 makes it a clear option for teams seeking a free library they can deploy themselves; the offer is a library, not a priced hosted OCR service.

Platforms

docTR is available for Linux, macOS, and Windows, as well as API and self-hosted use. Docker images support deployment in containerized environments. Its primary platform is desktop, but its library and API template are aimed at development and deployment workflows rather than an end-user desktop interface.

Who it's for

Choose docTR if you are a developer or researcher who wants a free, customizable OCR library, needs layout detection alongside text recognition, or intends to train models for a particular document type. Look elsewhere if you need a hosted service, a conventional ready-to-use OCR app, or must run on an older Python version. Community questions can go to GitHub Discussions or the #doctr Slack channel; the project also welcomes issues, pull requests, and community models.

Pros and cons

• Pro: Text recognition, layout prediction, and custom training sit in one library, which suits workflows that need structured document understanding.
• Pro: Free access, multiple installation routes, and CPU/GPU inference options leave deployment choices to the user.
• Pro: Text and JSON-serializable outputs make it practical to feed extracted content into other software.
• Con: Python 3.11 or higher is mandatory, limiting use in older Python environments.
• Con: Training custom models and deploying the REST template require development and operational effort; docTR is not presented as a hosted OCR service.
• Con: HTML input needs an optional extra, and the additional artefact detector depends on onnxruntime.

Alternatives

Pick PaddleOCR if mobile platforms matter: it lists Android and iOS support and an Official API Free Tier with a daily parsing limit of 20,000 pages. Choose OCRmyPDF for free self-hosted software that depends on external OCR and PDF tools. EasyOCR is another free option for Linux, macOS, and self-hosted use.

Consider FreeOCR.AI if you want a freemium option with web support: its Monthly Pro plan is $10 a month, billed monthly, for 10,000 credits per month, and its Yearly Pro plan is $80 a year, billed annually, for 120,000 credits per year. i2OCR is a web alternative whose free plan processes one image or PDF page at a time. Tesseract OCR is a free alternative supporting Linux, Windows, macOS, and Android. Scribe OCR offers a free, open-source web application alongside self-hosted use. Prizmo is a freemium option for iOS and macOS, with on-device OCR and cloud OCR plans.

For broader comparisons, browse OCR software or OCR API Software.

Verdict

docTR is a strong fit for developers and researchers who want free OCR they can embed, extend, and deploy themselves.