Alt Text Generator

A Docker image that automatically generates alt text for all Figures in PDF document.

Table of Contents

• Alt Text Generator
  • Table of Contents
  • Getting Started
  • Exporting Configuration for Integration
  • License
  • Help & Support

Getting Started

To use this Docker application, you'll need to have Docker installed on your system. If Docker is not installed, please follow the instructions on the official Docker website to install it. First run will pull the docker image, which may take some time. Make your own image for more advanced use.

Run using Command Line Interface

To run docker container as CLI you should share the folder with PDF to process using -v parameter. In this example all Figure tags without alt text will get description of image.

docker run -v $(pwd):/data -w /data --rm pdfix/alt-text-vision:latest generate-alt-text -i input.pdf -o output.pdf --model /model

With PDFix License add these arguments.

--name ${LICENSE_NAME} --key ${LICENSE_KEY}

Argument --model /model is required in order to tell vision where model is located. Without this vision would fail to find model.

You can also generate alt text just for one image using this example:

docker run -v $(pwd):/data -w /data --rm pdfix/alt-text-vision:latest generate-alt-text -i image.jpg -o output.txt --model /model

For more detailed information about the available command-line arguments, you can run the following command:

docker run --rm pdfix/alt-text-vision:latest --help

Exporting Configuration for Integration

To export the configuration JSON file, use the following command:

docker run -v $(pwd):/data -w /data --rm pdfix/alt-text-vision:latest config -o config.json

License

This project uses the vit-gpt2-image-captioning model provided by nlpconnect via Hugging Face, which is licensed under the Apache 2.0 License. See THIRD_PARTY_LICENSES.md for details.

The Docker image includes:

• Vision + GPT2 Image Captioning, Apache‑2.0 License
• PDFix SDK, subject to PDFix Terms

Trial version of the PDFix SDK may apply a watermark on the page and redact random parts of the PDF including the scanned image in background. Contact us to get an evaluation or production license.

Help & Support

To obtain a PDFix SDK license or report an issue please contact us at support@pdfix.net. For more information visit https://pdfix.net