Cross-Modality Feature Alignment

Learning Knowledge from Textual Descriptions for 3D Human Pose Estimation
Mainstream 3D human pose estimation methods directly predict 3D coordinates of joints from 2D keypoints, suffering from severe depth …
Cite