This is a repository for the openpose component of Edusense.
This instruction is for manually setting up openpose pipeline. For those who are new to Edusense, we recommend to take a look at automated multi-container setup provided here or use an automated script provided here.
Note-: Openpose pipeline processes the video, extracts the keypoints and sends them to the video pipeline via either local unix socket (demonstrated below, remember to use the same log directory for both openpose and video pipeline) or TCP socket
docker build .
You may also want to name the image like below:
docker build . -t <tag_name>
Make a log directory (log_dir) and pass the path to the docker run command. Make sure, to have a video in a video directory,pass the directory path and the video name to the docker run command.
nvidia-docker run \
--name <unique name for container> \
--rm \
-e LOCAL_USER_ID=$(id -u) \
-v <log_dir: path of the log dir>:/tmp \
-v <video_dir: path of the video directory>:/app/video \
<image name for the container/tag_name with which the image is built>\
--video /app/video/<videoname = name of the video> \
--num_gpu_start <GPU number to be assigned to the container> \
--num_gpu <Number of GPU assigned to the container(preferred 1)> \
--use_unix_socket \
--unix_socket /tmp/unix.front.sock \
--display 0 \
--render_pose 0 \
--raw_image \
--process_real_time \
Note-: For real-time processing, pass the RTSP URL to ip_camera argument
When compiling, we inject our EduSense code into OpenPose code base. OpenPose is a library under active development; our code often fails to compile because of code changes (e.g., API and data structure) in newer versions of OpenPose. We need to update our code whenever we integrate newer OpenPose into EduSense.
We release our EduSense code as patch diff files. The original source code we derive the current patch files from is /openpose/examples/tutorial_api_cpp/16_synchronous_custom_output.cpp . The file names of the code changes time to time, so we need to track file changes across OpenPose versions.
To generate a diff patch file,
diff <original file> <new edusense file> > edusense/edusense.cpp.patch
Once the patch file is ready, we also need to modify Dockerfile to update the path to the original source file. A new patch file lacks license notices, so we need to update the patch file also.
To apply a patch to a file:
patch <original file> -i examples/edusense/edusense.cpp.patch -o examples/edusense/edusense.cpp
If a developer also wants to update CMakefile, he/she can follow the same instructions noted above.