Giters

hanquansanren / docUnet.pytorch

This is a pytorch implementation of DocUNet: Document Image Unwarping via A Stacked U-Net

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

DocUNet: Document Image Unwarping via A Stacked U-Net

paper zhihu

how to run

  1. prepare some scans and background images
  2. use flow scipt to generated some images and labels
python3 data_generator/generator.py -i scan_images_path -b background_images_path -o output_path
  1. change trainroot in config.py
  2. use flow scipt to train
python3 train.py

how to eval

  1. change model_path and img_path in predict.py
  2. use flow scipt to train
python3 predict.py

generated images sample

data

result

result on training data

left：input, right output result1

result on really data

left：output, right input result1

About

This is a pytorch implementation of DocUNet: Document Image Unwarping via A Stacked U-Net

Languages

Language:Jupyter Notebook 57.4%Language:Python 19.5%Language:C++ 11.0%Language:Makefile 4.8%Language:C 3.9%Language:CMake 3.4%