OpenPose has represented the first real-time multi-person system to jointly detect human body, hand, facial, and foot keypoints (in total 135 keypoints) on single images.
The install guide on its website, it starts with dependencies that should be installed before installing openpose. “DO NOT FOLLOW IT!”. Instead, follow these steps:
1) check your Nvidia driver version in additional drivers page on your ubuntu. Make sure it is one of the proprietary packages.
nvcc --version
If CUDA is not installed, install nvidia toolkit using
sudo apt-get install nvidia-cuda-toolkit
and reboot.
3) Download a version of cuDNN from here that matches your CUDA version. Download all the .deb files for runtime and developer modes and code samples.
cp -r /usr/src/cudnn_samples_v8/ $HOME
cd $HOME/cudnn_samples_v8/mnistCUDNN
make clean && make
chmod +x ./mnistCUDNN
If cuDNN is properly installed and running on your Linux system, you will see a message saying test passed!
5) Install opencv
sudo apt install libopencv-dev python3-opencv
6) Install and configure gcc version 8 as the default using info from this link . Check the gcc version and make sure it is version 8.
7)Now, clone openpose from the github
git clone https://github.com/CMU-Perceptual-Computing-Lab/openpose
8) Install OpenPose dependencies by running the shell script:
cd openpose\scripts\ubuntu
chmod +x install_deps.sh
sudo ./install_deps.sh
This will install all the dependencies except caffe. caffe source code is already in openpose 3rdparty folder. When you build openpose, caffe will be installed.
9) Configure and make the build for OpenPose:
cd {OpenPose_folder}
mkdir build/
cd build/
cmake-gui ..
select the configs you want. Make sure to check build caffe, use cuDNN and openCV. Hit ‘Configure’ and then ‘Generate’. Then close cmake.
10) Build OpenPose
cd build/
make -j5
11) Now, you can run a demo using the following code:
cd ..
./build/examples/openpose/openpose.bin --video examples/media/video.avi
