Skip to content

Latest commit

 

History

History
130 lines (102 loc) · 4.64 KB

README_en.md

File metadata and controls

130 lines (102 loc) · 4.64 KB
Raw

📊RapidUnWrap

SemVer2.0 GitHub

Recent Updates

  • 2024.11.15
    • Completed the initial version of the code, converted the UVDoc model to onnx, and improved pre- and post-processing.
  • 2024.12.15
    • Added deblurring/shadow removal/binarization functions and models, upgraded to RapidUndistort.

Introduction

This repository is used for correcting document distortion, deblurring documents, shadow removal, and document binarization. It provides multiple models and flexible task combinations, supports automatic model downloading. Original PyTorch model sources can be found in the Acknowledgments section. Quick Start Usage Suggestions Parameter Explanation Model Address

Online Demo

modelscope huggingface

Effect Showcase

res_show.jpg res_show1.jpg

Installation

pip install rapid-undistorted

Quick Start

import cv2

from rapid_undistorted.inference import InferenceEngine
img_path = "img/demo.jpg"
engine = InferenceEngine()
# Distortion correction -> Shadow removal -> Deblurring (specify deblurring model)
output_img, elapse = engine(img_path, ["unwrap", "unshadow", ("unblur", "OpenCvBilateral")])
# Shadow removal -> Deblurring (specify deblurring model)
#output_img, elapse = engine(img_path, ["unshadow", ("unblur", "OpenCvBilateral")])
# Default selection of the first unblur model in the yaml configuration file
#output_img, elapse = engine(img_path, ["unshadow", "unblur"])
# Binarization as an alternative to shadow removal method
#output_img, elapse = engine(img_path, ["unwrap", "binarize", "unblur"])
print(f"doc unwrap elapse:{elapse}")
cv2.imwrite("result.png", output_img)

Usage Suggestions

  • For English and numeric deblurring, the NAFDPM model is better, but for Chinese text, using the OpenCV method is more suitable.
  • The shadow removal model has richer functionality and better results compared to binarization, so it is not recommended to directly use the binarization method.

Parameter Explanation

Initialization Parameters

Supports passing a config configuration file to declare the required task types and corresponding models, as well as paths.

config_path = "configs/config.yaml"
engine = InferenceEngine(config_path)
tasks:
  unwrap:
    models:
      - type: "UVDoc"
        path:
        use_cuda: false
        
  unshadow:
    models:
      - type: "GCDnet"
        sub_models:
          - type: "GCDnet"
            path:
            use_cuda: false
            use_dml: false
          - type: "DRnet"
            path:
            use_cuda: false

  binarize:
    models:
      - type: "UnetCnn"
        path:
        use_cuda: false

  unblur:
    models:
      - type: "OpenCvBilateral"
        path:
      - type: "NAFDPM"
        path:
        use_cuda: false

Execution Parameters

engine(img_path, task_list)
engine(img_path, ["unwrap", "unshadow", ("unblur", "OpenCvBilateral")])

Acknowledgments

unwrap: UVDoc unshadow: GCDnet unblur: NAFDPM binarize: UnetCnn

Contribution Guidelines

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

If you have other good suggestions or integration scenarios, the author will actively respond and support them.

Open Source License

This project is licensed under the Apache 2.0 license.