| { |
| "field_id": "F001", |
| "field": "3d human pose estimation", |
| "directions": [ |
| { |
| "direction_id": "G0", |
| "name": "Articulated human recovery from visual observations", |
| "scope": "Infer 3D human joints, pose, motion, or body shape from 2D pose, images, video, or contemporaneous multi-view observations; this also includes candidate poses and probabilistic priors under ambiguity." |
| }, |
| { |
| "direction_id": "G1", |
| "name": "Context-constrained human reconstruction", |
| "scope": "When relationships between the human body and other people, the camera, the environment, or unseen body regions determine whether a reconstruction is credible, jointly infer the human body and its location, motion, or constrained hidden parts." |
| }, |
| { |
| "direction_id": "G2", |
| "name": "Body recovery from partial and specialized measurements", |
| "scope": "Recover full-body pose, motion, or shape from self-mounted observations that do not directly show the complete body, or from directly acquired 3D point measurements that may be sparse or incomplete." |
| }, |
| { |
| "direction_id": "G3", |
| "name": "Training evidence and transferable representations for human recovery", |
| "scope": "When direct 3D annotations are limited, training coverage is insufficient, or downstream tasks require transferable representations, construct supervision, training examples, or pretrained representations to support pose and body recovery." |
| }, |
| { |
| "direction_id": "G4", |
| "name": "Human reconstruction under computational constraints", |
| "scope": "Maintain usable 3D pose or mesh reconstruction quality while reducing model size, inference computation, or device burden." |
| } |
| ] |
| } |
|
|
