This docker file is built on top of the caffe gpu image. We rebuild caffe, knowing we have the correct requirements, as for some reason openpose does not find all the includes. The image includes all the available models at this time. Made available on Docker.
The image can be built with:
sudo docker build -t openpose:gpu .
There is now an image on Docker hub.
docker pull dgrnwd/openpose-gpu
Once the image is built we connect a data directory and run the image.
docker run --runtime=nvidia -v /media/data1:/data -it openpose:gpu bash
The openpose binary is in the path, so we can call it:
openpose -helpshort
The output from openpose -help is saved in help.txt
An example for tracking a face might be:
openpose \
-display=0 \
-video /data/small-test.mov \
-face=true \
-face_net_resolution="368x368" \
-net_resolution="-1x640" \
-write_json="/data/op_out/kp_368_640" \
-write_video="/data/op_out/face_368_net_640.avi" \
-camera_fps=59.94
Another example, from the image pulled from docker hub.
sudo docker run --runtime=nvidia \
-v ~/emo-data:/data \
-it dgrnwd/openpose-gpu:latest bash
openpose \
-display=0 \
-video /data/aj-2k-sample.mov \
-frame_rotate=270 \
-face=true \
-write_json="/data/op_out/kp-aj-2k-sample-default" \
-write_video="/data/op_out/kp-aj-2k-sample-default.avi" \
-camera_fps=59.94