This is an implementation of the waping process presented in Chapter 3 in the paper "DocUNet: Document Image Unwarping via a Stacked U-Net".
@InProceedings{Ma_2018_CVPR,
author = {Ma, Ke and Shu, Zhixin and Bai, Xue and Wang, Jue and Samaras, Dimitris},
title = {DocUNet: Document Image Unwarping via a Stacked U-Net},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2018}
}
This code was produced with the help of:
- https://stackoverflow.com/questions/53907633/how-to-warp-an-image-using-deformed-mesh
- mhashas/Document-Image-Unwarping-pytorch#7
This code aims for a better variable naming and better structure.
First, you must install the required libraries with the command
pip install -r requirements.txt
Then you can run the warper with the following command:
python src/warper.py <path_src_img> <name_dst_img>.jpg
Once you run the command, the warped image will be saved in the assets folder
with the name <name_dst_img>.jpg.
The parameters of the algorithm can be changed in the config.json file. The
parameters are:
max_strength: defines the strength of the modification. The higher the parameter, the stronger is the deformation.max_random_mov: defines the number of deformations applied to the image.curved_normalizer: defines a normalizer factor to the curved folding. The The higher the parameter, the smaller the effect of the curved folding.resize_factor: defines the length of the additional margin added to the original image to prevent the corners of the paper to be outside of the warped version. The smaller the paremeter, the thicker the margins.
With the current implementation, the hyperfine command shows the following
times for an image with dimensions (1776, 1200):
Time (mean ± σ): 3.904 s ± 1.948 s [User: 3.541 s, System: 1.017 s]
Range (min … max): 1.157 s … 6.252 s 10 runs