DocScanner: Document Image Rectification

This Space demonstrates DocScanner, a deep learning model that automatically corrects geometric distortions in document images.

If you have a photo of a document that is warped, skewed, or has curled edges, this tool can transform it into a flat, top-down, scanner-like image.

This application is an implementation of the research paper: DocScanner: Robust Document Image Rectification with Progressive Learning (https://arxiv.org/abs/2110.14968).

How to Use

  1. Upload an Image: Drag and drop a distorted document image into the input box, or click to browse your files.
  2. Submit: Click the "Submit" button to begin the rectification process.
  3. View the Result: The corrected, flattened document will appear in the output box on the right.

Technical Details

  • Model: This demo uses the DocScanner-L model, as described in the paper.
  • Technology: The application is built with Python, PyTorch, and the Gradio library.
Examples