UTE

UTE (Universal Translation Engine) is a tool for translating images and documents into a different language. It is similar to popular phone apps in that it overlays the translated text in the correct place. All of the pieces of desktop software I have tried put the translation in a separate box or do not properly overlay it onto the original image, making it hard to understand the structure of the document. Unlike these phone apps, UTE is an open-source, embeddable engine, allowing it to be used in other tools.

How it works

The pipeline is essentially: tesseract (OCR) -> Deepl API -> annotate image/document. Other translation services to come.

Usage

Download the repo and install dependencies (proper packaging to come):

git clone https://github.com/MarcelRobitaille/ute.git
cd ute
pip install -r requirements.txt

Currently the command-line interface supports two commands.

python -m ute.ute translate-image <Source> <Destination>
python -m ute.ute translate-pdf <Source> <Destination>

Starting the REST API:

fastapi dev ute --app fastapi_app

Roadmap

• Engine and basic CLI
• Rest API
• Integration into Paperless-ngx
• Other translation engines (like libretranslate)
• Screen overlay to translate a region of your screen, like taking a screenshot
• Live screen overlay. Like above, but live instead of one time. This will probably require major performance improvements and diffing of the changed regions of the image

Other todos

UTE is a work in progress. For example, it only works with pre-straightened (de-skewed) images. This section tracks some general improvements that do not really fit as roadmap items.

• Performance (profiling or rewrite in not python)
• Better config handling
• Straighten (de-skew)
• Packaging in package repositories