| { |
| "field_id": "F008", |
| "field": "6d object pose estimation", |
| "directions": [ |
| { |
| "direction_id": "G0", |
| "name": "Pose generalization to unseen objects", |
| "scope": "Estimate the poses of objects unseen during training, including methods that use target-specific reference images or representations and methods that use category representations to infer pose, and where relevant size and shape, across instances." |
| }, |
| { |
| "direction_id": "G1", |
| "name": "Pose recovery by observation–geometry alignment", |
| "scope": "Solve or refine object pose using constraints between observations and object geometry or candidate poses. This includes local correspondences, render-based comparison, and joint multi-view constraints; a method need not use all of these approaches." |
| }, |
| { |
| "direction_id": "G2", |
| "name": "Pose learning with reduced real-label dependence", |
| "scope": "Obtain usable pose estimators when manually annotated poses in real images are scarce, using synthetic or self-supervised training evidence or adaptation with unlabeled target-domain data." |
| }, |
| { |
| "direction_id": "G3", |
| "name": "Temporal object-pose tracking", |
| "scope": "Maintain an object's 6D pose across successive observations while handling motion, unreliable observations, or recovery after tracking failure." |
| }, |
| { |
| "direction_id": "G4", |
| "name": "Structured pose representation and uncertainty", |
| "scope": "Represent the geometry of rotations and ambiguity in observations appropriately in pose prediction or inference, including structured rotation features, multiple plausible poses, pose distributions, or uncertainty that can be propagated." |
| }, |
| { |
| "direction_id": "G5", |
| "name": "Joint multi-object pose prediction", |
| "scope": "Jointly localize or distinguish multiple object instances in the same scene and predict each instance's pose, rather than only improving pose recovery for objects already localized individually." |
| }, |
| { |
| "direction_id": "G6", |
| "name": "Pose data and evaluation infrastructure", |
| "scope": "Create data, annotations, or evaluation protocols for object-pose research, with reusable research and evaluation infrastructure as the central contribution." |
| } |
| ] |
| } |
|
|
