SRCNN_Cpp

This project is an open source project of "C++ Implementation of Super-Resolution resizing with Convolutional Neural Network".

latest update

• Pull request accepted from SRCNN_OpenCV_GCC, which has been verified in Linux environment with updated OpenCV 4. Original reporitory with OpenCV 3 has been archived in V1.0.0 branch. Thanks for the contributions from rageworx.
• OpenCV 4 compatibility updated.
• Supporting macOS universal binary.

Non-OpenCV models

OpenCV is too large for specific applications, you may need the following two non-OpenCV models in some cases.

• libsrcnn is a stand-alone non-OpenCV library, with 4 times more faster than OpenCV model: rageworx/libsrcnn.
• stb-image-srcnn is an stb-based non-OpenCV library: zvezdochiot/stb-image-srcnn.

Introduction

This is an open source project from original of this: SRCNN_Cpp is a C++ Implementation of Image Super-Resolution using SRCNN which is proposed by Chao Dong in 2014.

• If you want to find the details of SRCNN algorithm, please read the paper:

  Chao Dong, Chen Change Loy, Kaiming He, Xiaoou Tang. Learning a Deep Convolutional Network for Image Super-Resolution, in Proceedings of European Conference on Computer Vision (ECCV), 2014

• If you want to download the training code(caffe) or test code(Matlab) for SRCNN, please open your browse and visit http://mmlab.ie.cuhk.edu.hk/projects/SRCNN.html for more details.

• And thank you very much for Chao's work in SRCNN.

What changed ?

1. Code modification from original SRCNN.
2. OpenMP parallel processing, improved performance.
3. Supports almost of platform - POSIX compatibled.
   • MSYS2 and MinGW-W64
   • GCC of Linux
   • LLVM or CLANG of macOS, suporting universal binary build.

License

The repo is released under the GPL v2 License (refer to the LICENSE file for details).

Example

• Origin:

  

• Bicubic interplation x1.5

  

• SRCNN x1.5

  

Requirements

1. Windows may need MSYS2 and MinGW-W64.
2. You need to install latest version of OpenCV to your build environments, install opencv libraries into your system with one of these:
   • MSYS2: pacman -S /mingw-w64-x86_64-opencv
   • Debian: sudo apt-get install libopencv-dev
   • MacOS
     1. Before install Brew :

     sudo xcode-select --install 
     sudo xcodebuild -license
     ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"

     1. After installed Brew :

     brew update
     brew install opencv4

3. Manual staic build OpenCV [Refer to OpenCV Installation]
   • Clone or download OpenCV source to you base level directory of this sources.

   cd ~/<my_working_directory>
   git clone https://github.com/opencv/opencv.git
   git clone https://github.com/opencv/opencv_contrib.git

   • Go to opencv, then make a 'build' directory.

   cd ~/opencv
   mkdir build
   cd build
   cmake -D CMAKE_BUILD_TYPE=Release -D CMAKE_INSTALL_PREFIX=/usr/local ..
   make -j8 # runs 8 jobs in parallel
   sudo make install

   • In case of MSYS2 Makefile, type like this.

   cmake -G "MSYS Makefiles" -DBUILD_SHARED_LIBS=OFF -DENABLE_PRECOMPILED_HEADERS=OFF -DWITH_IPP=OFF -DWITH_TBB=OFF -DWITH_FFMPEG=OFF -DWITH_MSMF=OFF -DWITH_VFW=OFF -DWITH_OPENMP=ON ..
   

   • This project doesn't using video decoding, and there's too many erorrs occurs on Video processing source in OpenCV ( damn sucks cmake options, they're useless )

Compile and Use

You can compile the C/C++ files on the command line in your POSIX shell.

make

If the compile is successful, you will see linked binary in 'bin' directory.

./bin/srcnn ./Pictures/test.jpg 

The generated image will be ./Pictures/test_resized.jpg.