SETUP for OPENPOSE and 3D-POSE-BASELINE on Ubuntu 16.04 / 180505

Note: OpenPose alone can record 3D coordinates with a multi-camera setup. However the 3d-pose-baseline method allows generating 3D coordinates for multiple subjects from a single camera.

1. Clone OpenPose:
   https://github.com/CMU-Perceptual-Computing-Lab/openpose
   

2. Install CUDA, CUDNN, and OpenCV

./ubuntu/install_cuda.sh
./ubuntu/install.cudnn.sh
sudo apt-get install libopencv-dev python-opencv

3. Download models

./models/getModels.sh

4. Install Caffe and build OpenPose

./ubuntu/install_caffe_and_openpose_if_cuda8.sh

5. Sample OpenPose shell scripts for Ubuntu and Windows.

6. Clone 3D-pose-baseline (fork with OpenPose converter script):
   https://github.com/ArashHosseini/3d-pose-baseline
   

7. Install Python dependencies:

pip install h5py
pip install tensorflow
pip install imageio
pip install matplotlib

8. Download models:
   

(Place in ./experiments/)
https://drive.google.com/file/d/0BxWzojlLp259MF9qSFpiVjl0cU0/view?usp=sharing

(Place in ./data/h36m/)
https://www.dropbox.com/s/e35qv3n6zlkouki/h36m.zip

9. Create a directory ./png for output.

10. Run a sample shell script.