Pytorch implementation and extension of "DocUnet: Document Image Unwarping via A Stacked U-Net"
This repository contains an unofficial implementation of DocUNet: Document Image Unwarping via a Stacked U-Net. We extend this work by:
Unfortunately, I am not allowed to make public the dataset. However, I created a very small toy dataset to give you an idea of how the network input should look. You can find this here. The idea is to create a 2D vector field to deform a flat input image. The deformed image is used as network input and the vector field is the network target.
conda env create -f environment.yml
conda activate unwarping_assignment
save_dir
command line argument.inference_dir
should be used to provide the
relative path to the folder which contains the images to be classified.