docTR
An Apache-2.0 Python OCR library on PyTorch that finds and reads words in PDFs and images, with optional layout detection and a key-information extraction predictor.
About docTR
Developers need OCR they can run and fine-tune themselves instead of sending documents to a hosted API.
Who it’s for
- Python developers building document pipelines
- Teams that want to self-host or train OCR models
When to consider it
Consider docTR when you want an open-source, two-stage OCR pipeline (text detection then recognition) with selectable pretrained models that you can install with pip and run on your own hardware.
Tradeoffs & limitations
- A Python library rather than a hosted service: it needs Python 3.11 or higher and PyTorch, and you run and scale it yourself. Created by Mindee and now developed and maintained by t2k GmbH.
SOTA overview · Documentation-based assessment · Sources & review method
Updates
No updates shared yet.
Discussion
Newest firstAsk a question or share how you use docTR.
Keep it helpful. Community rules
Loading discussion…
Sources & review method
Documentation-based assessment · Oct 31, 2026 · Prepared with AI assistance; not a hands-on benchmark.
Checked by SOTA · AI-assisted documentation review. Selection advice is our assessment; verify current requirements for your deployment.
Import history & original evidence
docTR on GitHub
An Apache-2.0 Python OCR library on PyTorch that finds and reads words in PDFs and images, with optional layout detection and a key-information extraction predictor.
Based on official pages and announcements checked on 2026-10-31. No hands-on test, performance benchmark or popularity ranking is claimed.