Fig. 1: We present a simple but
highly effective framework for adapting human pose
estimation methods to highly truncated settings
that requires no additional pose annotation.
We evaluate the approach on
by annotating four Internet video