It covers the basis for creating a functional library that uses CUDA and OpenCV. In later stages, we show how to use this library in an executable. The project takes advantage of modern CMake syntax and CUDA.
Start by clonning this repository from https://github.com/LuisTbx/SkinCuda
Before compiling, create a test_images directory and place some images containing people with different skin colors. You can also download some videos such as:
mkdir test_images
cd test_images
wget -O test_images/test_image.jpg https://assets.weforum.org/article/image/XaHpf_z51huQS_JPHs-jkPhBp0dLlxFJwt-sPLpGJB0.jpg
cd ../
We use CMake to build this project. Let’s create now a build directory, then run cmake
mkdir build
cd build
cmake ../
You may find issues about the GPU architecture, please adjust accordingly on lines 30, 44, and 58 of the CMakeLists.txt file. As it is it is mean to work with the 10XX Pascal series compute capability 61.
set_target_properties(${SUBPROJECT} PROPERTIES CUDA_ARCHITECTURES "61")
At this point we are ready to buld the project, use make to start building. Once built you will find three executables, skin processes images from the webcam or vide, skinMap crates the skin map on a single image and skinMask produces the binay mask. After running the executables will produce skinMap.jpg and skinMask.jpg images.
make -j6
./skin # Process webcam
./skin test_images/video.avi # Process video file
./skinMap test_images/test_image.jpg
./skinMask test_images/test_image.jpg
ls
The skinnMap image is the result of applying the algorithm to the original image, it removes the background by blacking out the values (0) on the original image. The skinMask is a binary mask where the skin pixels are set to 255 and the background 0.
And that is it, hope you manage to reproduce these steps. If you have any suggestion or issue please reach out by crating an issue or a PR on GitHub!