TGV Image Denoising

This code repository is an implementation of the total generalized variation method based on this paper

Platforms

• Linux (Tested)

Requirements

These are the base requirements to build

• Cmake
• A C++17-standard-compliant compiler
• For GPU version you actually need something, that support OpenCL 1.2 (so basically its not always GPU)

Installing

  mkdir build
  cd build
  cmake -D BUILD_RELEASE:BOOL=true ..
  cmake --build .

Running the tests

Compile with Cmake flag -D BUILD_RELEASE:BOOL=false

  mkdir build
  cd build
  cmake -D BUILD_RELEASE:BOOL=false ..
  cmake --build .

How to use

Current solution based on command line interface with keys.

| Key | Purpose | Default Value | | :------------------- | :-------------------------------------- | :------------ | | -c | Use CPU | False | | -g | Use GPU | True | | -n <Num> | Amount of Iterations | 1000 | | -p <Folder Path> | Path to folder with data | data | | -a <Num> | Index of gpu if there are a more than 1 | 0 | | -r <File name> | Name of the result file | result | | -i <Num> | amount of images for GPU | 10 | | -scaleX <Num> | scale for X axis for Ply file | 1.0 | | -scaleY <Num> | scale for Y axis for Ply file | 1.0 |

So just start program with these keys

File format

Program works with PFM, so perhaps you will have to convert files

Examples

./TGV -g -n 400 -p "data" -a "0" -r "resultFileName" -i 14

./TGV -p "data" -a "0" -r "DenoisedImage" -i 6

Authors

• Daniil Smolyakov - Initial work and CPU/GPU based code - DanonOfficial

License

This project is licensed under the MIT License - see the LICENSE.md file for details