Skip to content

Pipelines overview

Skeleton pipeline#

The overall skeleton pipeline goes like so

fullpipe.png

But note these steps can be left out so we can dump first and do tracking later if we like e.g. first run

dumppipe.png

And then later run

filterpipe.png

Pipelines starting with the pose estimator are run using the dump command, while pipelines starting from existing pose dumps using the filter command. Which method is used for different . See next CLI examples and CLI reference.

Face pipeline#

The face pipeline can run in two modes. In the first mode, which is not recommended for most usages, dlib's face detection and face keypoint detection pipeline is used. There is some information about the dlib keypoint detection in this blog post. In the second mode, an existing pose dump including these 68-keypoints as estimated by OpenPose is used. The second is preferred in most situations.

Dlib only pipeline:

dlibpipe.png

Skeleton-based pipeline:

dlibpipe.png