The project is tested on Windows and Linux (with all necessary libraries pre-installed manually)

Project structure:

• dataset -> the directory with test images
• dll -> the directory with dianmic libraries
• facedetector -> the source directory for the face search function
• helpers -> the source directory for the helper classes
• include -> the directory with include files
• test_image -> the directory with test images for Gtest
• weights -> the directory with weights of neural network

List of required libraries:

• OpenCV
• Boost
• Gtest

For windows, the download of the necessary libraries can be done in Vcpkg (https://github.com/microsoft/vcpkg). Below is the script for automatic installation of all dependencies.

    > git clone https://github.com/microsoft/vcpkg
    > cd vcpkg
    > .\vcpkg\bootstrap-vcpkg.bat
    > vcpkg install opencv:x64-windows
    > vcpkg install boost:x64-windows
    > vcpkg install gtest:x64-windows

There is no script for installing all dependencies for Linux. Unfortunately, I don't know Linux that well.

Architectural solutions in the project:

• From the condition that we do not know the number of images, it was decided to process them with batches of size 20.

• Jason file format:

   {
   "image_id": "../dataset\1\2\2\i (1).jpg",
   "faces": [ [281, 143, 594, 575] ]
   }
  

Where:

• "image_id" -> The relative path to the image from the folder
• "faces" -> An array of rectangles with the coordinates of the found faces. Coordinates in XYXY format