This project provides Python modules for the OpenPose in PyTorch project. Two types of components are implemented in C++ to accelerate the running efficiency and engineering deployment:

• Preprocessor: Converting the original keypoint coordinate label into two types (the Gaussian heatmap for keypoints and the heatmap for part affinity fields) of feature labels.
• Postprocessor: Analyzing the feature maps obtained by the deep neural network (DNN) to get the keypoint coordinates of multiple objects (usually person).

If you are benefited from this project, a donation will be appreciated (via PayPal, 微信支付 or 支付宝).

• Linux is recommended, and Windows is not tested.
• Compiler supports C++11 or later (such as GCC 4.8 or later).
• Python development library, version 3 is recommended.
• CMake.
• pybind11.
• Boost.
• Eigen3.
• OpenCV3.
• cnpy (optional).

• Install the depend tools and libraries.
• Download and extract the source code (e.g., ~/code/pyopenpose).
• Create the build directory (e.g., ~/code/pyopenpose-build) and generate CMake cache (mkdir ~/code/pyopenpose-build && cd ~/code/pyopenpose-build && cmake ~/code/pyopenpose)
• Check if the following variables in ~/code/pyopenpose-build/CMakeCache.txt are correct:
  • PYMODULE_ROOT: should be the user location for Python modules (e.g., ~/.local/lib/python3.5/site-packages)
  • PYTHON_EXECUTABLE and PYTHON_LIBRARY: should be the interpreter and library of the same version of Python. Version 3 is recommended (e.g., /usr/bin/python3 and /usr/lib/x86_64-linux-gnu/libpython3.5m.so).
  • C++11 is enabled (the CMAKE_CXX_FLAGS is set to -std=c++11 -fPIC by default).
  • DEBUG_SHOW can be used to debug the program in test projects (in the test folder of the source directory).
• Compile the source code (cd ~/code/pyopenpose-build && make).

If everything goes well, a dynamic library (e.g., ~/.local/lib/python3.5/site-packages/pyopenpose.cpython-35m-x86_64-linux-gnu.so) which contains the Python modules will be appeared and can be used in Python codes.