/OpenPoseTutorial

Installing and Running OpenPose Docker on SSRDE

Setting Up OpenPose Environment

  1. Log into SSRDE and go to the path where you want to create and run Openpose
  2. Create a directory for Openpose. In this example, the directory is called "openpose_testing" so to make the directory, enter the command mkdir openpose_testing
  3. Go inside the directory with cd openpose_testing, and enter the command: singularity pull docker://szreik/openpose-cpu-face:latest
  4. When completed, you should see a file named "openpose-cpu-face_latest.sif" in your directory. It can be renamed to whatever you like as long as it has .sif extension
  5. Create a new directory to store input videos. In this example, the directory is called "videodata" so enter the command mkdir videodata. Make sure you are in the openpose_testing directory when you enter this command.
  6. Go inside the videodata directory with the command cd videodata, and make a directory named "json" by entering mkdir json.

Uploading videos

You can upload videos one of two ways:

  • On Mac, you can connect to the server through Finder so you can visually see the files and drag them from your local machine to SSRDE. How to Connect to Remote Server on Mac

    • Drag the video from your local machine into your videodata directory on SSRDE

  • Using the scp command:

    • On you local machine (not in SSRDE!) go to the directory where you have your video, and enter the command scp {NAME OF VIDEO FILE} {SSRDE USERNAME}@ssrde.ucsd.edu:/{PATH TO STORE VIDEO ON SSRDE}. Once it fully uploads you should be able to find the video in the SSRDE path you inputted.

    • Using the example path names we created above, and considering that I am storing my "openpose_testing" directory in /sphere/greene-lab/lab_members/salma/, the scp command would look like this: scp video.mp4 szreik@ssrde.ucsd.edu:/sphere/greene-lab/lab_members/salma/openpose_testing/videodata

Running OpenPose on Videos

  1. Now that our video is on SSRDE, we can send a job to SSRDE to run Openpose!

    To do this, you need a shell script. The current OpenPose script can be found in /sphere/greene-lab/lab_members/salma/openpose_testing and is named "run_openpose.sh". The command to run OpenPose from my directories is already there, but if you created these directories somewhere else you can make a copy of the script and replace the command like so: singularity exec -B {NAME OF FULL PATH TO VIDEO DATA DIRECTORY}:/data {NAME OF DOCKER IMAGE FILE} \ /openpose/build/examples/openpose/openpose.bin --model_folder /openpose/models \ —number_people_max 1 --render_pose 0 --face --face_render 1 \ --video /data/$1.mp4 --display 0 \ --write_video /data/$1_openpose.mp4 --keypoint_scale 4 \ --write_json /data/json/$1

  2. To queue the job on SSRDE, you need to enter the following command: sbatch -J {NAME OF JOB} -p general --mail-user={YOUR EMAIL} --mail-type=ALL -o {PATH TO OPENPOSE TESTING DIRECTORY}/JOB NAME}_output.o%j.txt {PATH TO OPENPOSE SCRIPT}/run_openpose.sh {NAME OF VIDEO FILE WITHOUT .mp4 EXTENSION}

    For me, the command will look like this if I am running it on a video named pt102.mp4: sbatch -J PT102 -p general --mail-user=szreik@ucsd.edu --mail-type=ALL -o /sphere/greene-lab/lab_members/salma/openpose_testing/PT102_output.o%j.txt /sphere/greene-lab/lab_members/salma/openpose_testing/run_openpose.sh pt102

You should now be able to enter the command squeue and see your job in the queue. You will get an email when the job starts/ends and you can find the output of running the script in your openpose_testing directory.

For all future videos, you now just have to upload them onto SSRDE and enter the sbatch command in step 2 of this section!