Abstract: Accurate acquisition of 3-D human joint poses holds significant implications for tasks such as human action recognition. Monocular single-frame 2-D -to-3-D pose estimation focuses on ...