Pointless Pose: Part Affinity Field-Based 3D Pose Estimation without Detecting Keypoints

Human pose estimation finds its application in an extremely wide domain and is therefore never pointless. We propose this paper a new approach that, unlike any prior one that we are aware of, bypasses the 2D keypoint detection step based on which 3D estimated, thus Our motivation rather straightforward: vulnerable to occlusions out-of-image absences, case errors propagate recovery deteriorate results. To end, resort explicitly estimating human body regions of interest (ROI) their orientations. Even if portion body, like lower arm, partially absent, predicted orientation vector pointing from upper arm will take advantage local image evidence recover pose. This achieved, specifically, by deforming skeleton-shaped puppet template fit estimated vectors. Despite simple nature, proposed yields truly robust state-of-the-art results several benchmarks in-the-wild data.