Netra OCR 1

An app for your own computer that reads Khmer and English documents and turns them into text you can correct and export.

InstallTry it in your browser
The Netra OCR editor: a scanned Khmer government notice on the left with its detected regions outlined, and the editable structured document on the right, with the same heading selected in both

Fewer mistakes than the alternatives

How many characters the app’s recognizer gets wrong on three Khmer test sets; lower is better. Qwen2.5-VL and DeepSeek-OCR were fine-tuned on the same training data first. Read the paper or see the model

KHOBThe public Khmer OCR benchmark: 325 real text lines
Legal documents220 lines from phone photos of certificates and ID cards
Printed words500 short words set in ten different fonts

Your documents never leave your computer

Nothing is uploaded

Pages are read on your machine, not sent to us or anyone else.

Works offline

After the one download, it doesn't need an internet connection.

Install it

Pick your system. It takes one command, and the app then opens in your browser like a website, but runs on your computer.

  1. 1

    Install Docker Desktop

    Download it from docker.com. In its settings, give it at least 4 GB of memory.
  2. 2

    Run this in PowerShell

    docker run -d --name netra-ocr -p 8000:8000 -v netra-data:/data ghcr.io/netra-ai-lab/netra-ocr
    The first run downloads about 260 MB.
  3. 3

    Open the app

    The first start takes a few seconds; the header says Ready · CPU when it’s done.Open localhost:8000

Questions

The page at localhost:8000 won't open

The app needs port 8000. The install command sets that up with -p 8000:8000. If you started the image from Docker Desktop’s Images tab instead, open Optional settings and set Host port to 8000.

After the first run, Netra OCR is listed under Containers in Docker Desktop, and you can start and stop it there without the terminal.

How do I update to a new version?

Run these three commands. Your documents are kept in the netra-data volume, so they survive the update.

docker pull ghcr.io/netra-ai-lab/netra-ocr
docker rm -f netra-ocr
docker run -d --name netra-ocr -p 8000:8000 -v netra-data:/data ghcr.io/netra-ai-lab/netra-ocr

What changed in each version is in the release notes.

Is there an API?

Yes, a REST API. Its reference opens at http://localhost:8000/docs while the app is running. The Python package is on PyPI.

Can I use it at work?

The Netra OCR code is released under the MIT licence. The Docker image also includes two components licensed under AGPL-3.0 (ultralytics and doclayout-yolo). The full list is in the third-party licences file; check it before using Netra OCR in a commercial setting.

Something isn't working

Tell us on GitHub issues, with what you ran and what you saw. The source code is there too.