| { |
| "field_id": "F006", |
| "field": "3d scene understanding", |
| "directions": [ |
| { |
| "direction_id": "G0", |
| "name": "Spatial semantic understanding of 3D scenes", |
| "scope": "Establish usable semantic assignments for locations or objects in 3D scenes, encompassing dense labels, distinct object regions, and spatial semantic representations that support flexible concept queries; these outcomes need not all be present together." |
| }, |
| { |
| "direction_id": "G1", |
| "name": "3D scene entity-relation modeling", |
| "scope": "Construct or predict structured graphs of entities and their spatial and semantic relationships from observed scenes." |
| }, |
| { |
| "direction_id": "G2", |
| "name": "Language-mediated 3D scene understanding", |
| "scope": "Align language with 3D scenes to locate entities referred to in language, or to answer questions, engage in dialogue, and understand questions about scenes; individual works need not output both entity locations and language answers." |
| }, |
| { |
| "direction_id": "G3", |
| "name": "Recovery and structured modeling of scene geometry", |
| "scope": "Infer the spatial form and organization of scenes from images or geometric observations, encompassing occupancy and semantic completion of unobserved regions, architectural layouts, depth and surface recovery, and scene models composed of objects or geometric primitives; these outputs are parallel branches." |
| }, |
| { |
| "direction_id": "G4", |
| "name": "Scene mapping across observations and change", |
| "scope": "Maintain a coherent spatial record of a scene through alignment, association, or memory updates across observations, and identify changes where applicable." |
| }, |
| { |
| "direction_id": "G5", |
| "name": "Transferable 3D scene feature learning", |
| "scope": "Learn features reusable for downstream 3D scene understanding tasks rather than targeting only one type of scene prediction output." |
| } |
| ] |
| } |
|
|
