Simple containerization of OpenPose for simple plug-and-play (aka OpenPose without tears).
You will need a host system with drivers for cuda 10
(1) clone this repository
git clone https://github.com/jutanke/openpose_docker.git
(2) build the docker container
cd openpose_docker && ./build.sh
(3) Usage:
./openpose.sh /your/img/input/dir /your/output/dir/keypoints/json
The keypoint json files will be located in the output folder.
If you want to extract the heatmaps and PAFs as well, simply call:
./heatmaps.sh /your/img/input/dir /your/output/dir/keypoints/json
The keypoints will be organized in the same way as with openpose.sh, however, you will find a directory heatmaps as subfolder to the output directory where the heatmaps and pafs are stored as single pngs. Please read the openpose documentation for a detailed structured definition.
If you just want to visualize your images using Openpose you can use the following script:
./visualize.sh /your/img/input/dir
Edit the path descriptions in torso_docker_run.sh
the first path should refer to the location of this folder the second should link to the location of the data-set. Run torso_docker_run.sh
. Within the container cd to /home/openpose_docker
and execute orientation_estimation.py
.