README.md

metadata

title: Source
app_file: demo.py
sdk: gradio
sdk_version: 3.40.1

Description

Due to work, more information will be released.

Performance

Execution Time when getting a 1024x960 corrected image.

Method	Time
TPS	-
Interpolation	-

Testing

1、Download model parameter and source codes

2、Resize the input image into 992x992

3、Running

• In GPU 0: python test.py --data_path_test=./your/test/data/path/ --parallel 0 --schema test --batch_size 1

• In CPU: python test.py --data_path_test=./your/test/data/path/ --parallel None --schema test --batch_size 1

Training

a) Download training data in here.

b) Run python train.py --data_path_train=./your/train/data/path/ --data_path_validate=./your/validate/data/path/ --data_path_test=./your/test/data/path/ --batch_size 32 --schema train --parallel 01

Use your Dataset

The training dataset can be synthesised using the scripts.

Q&A

1. Q:How to adjust the output image resolution？ A: Adjusting the ‘flat_shap’ or 'output_shape' in utilsV4.py. https://github.com/gwxie/Document-Dewarping-with-Control-Points/blob/4c74853b0eb93f7c6006a774c2eb42c64f363531/Source/utilsV4.py#L99 or https://github.com/gwxie/Document-Dewarping-with-Control-Points/blob/4c74853b0eb93f7c6006a774c2eb42c64f363531/Source/utilsV4.py#L162

2. Q:How to open '*.gw' files? https://github.com/gwxie/Document-Dewarping-with-Control-Points/blob/0f4e9ac577fb001a719fb63b05cfa915fe3c9866/Source/dataloader.py#L146 'im' is input images; 'lbl' is control points; 'segment' is the intervals of points between the horizontal and vertical directions;

3. Q:How to train the model again with 61 points? https://github.com/gwxie/Document-Dewarping-with-Control-Points/blob/1ad92be2995ee7f4ecdd04157350f2b44ecbd7e9/Source/dataloader.py#L185-L186 https://github.com/gwxie/Document-Dewarping-with-Control-Points/blob/1ad92be2995ee7f4ecdd04157350f2b44ecbd7e9/Source/dataloader.py#L74-L75 set self.col_gap = 0; self.row_gap = 0