| { |
| "field_id": "F002", |
| "field": "3d human reconstruction", |
| "directions": [ |
| { |
| "direction_id": "G0", |
| "name": "Recovery of observed human form and motion", |
| "scope": "Recover the 3D form, pose, or motion of a specific person from visual observations, including parametric bodies and meshes, static faces and heads, complete high-detail clothed surfaces, and reconstruction constrained by contact with other people, objects, or scenes." |
| }, |
| { |
| "direction_id": "G1", |
| "name": "Recovery of human surface appearance", |
| "scope": "Infer surface appearance usable for 3D rendering from limited observations of a person, including identity-consistent completion of color or texture on unseen surfaces and joint estimation of geometry, materials, and lighting to support changes in illumination." |
| }, |
| { |
| "direction_id": "G2", |
| "name": "Reconstruction of controllable clothing", |
| "scope": "Obtain clothing that can be treated as an independent object, and its relationship to the body, from observations of a person. This includes separating replaceable body and garment layers and recovering garment surfaces and deformation behavior as pose or time changes." |
| }, |
| { |
| "direction_id": "G3", |
| "name": "Renderable avatars of observed identities", |
| "scope": "Turn observations of identified bodies, hands, or heads into persistent, renderable 3D avatars that support novel views and, where applicable, new poses, expressions, or motions. This includes both personalized capture of recorded individuals and shared models that infer previously unseen individuals from sparse inputs." |
| }, |
| { |
| "direction_id": "G4", |
| "name": "Creation of new 3D human identities", |
| "scope": "Create 3D human-body or head identities and assets rather than recover observed people. This includes learning reusable population-level generative distributions and constructing specific renderable characters from text, pose, or other creation guidance." |
| } |
| ] |
| } |
|
|
