Applying Annotations to Three-Dimensional (3D) Object Data Models Based on Object Parts

摘要

Methods and systems for interacting with multiple three-dimensional (3D) object data models are provided. An example method may involve receiving a plurality of three-dimensional (3D) object data models that represent objects associated with a given category. The plurality of 3D object data models may include one or more annotated parts. The method may also involve receiving a candidate 3D object data model that is associated with the given category of objects. The candidate 3D object data model may include one or more un-annotated parts. The method may additionally involve, based on a shape and an appearance of the one or more annotated parts and spatial-layout information that defines how the one or more annotated parts are spatially arranged, applying an annotation of a respective part of a respective 3D object data model to a corresponding part of the candidate 3D object data model.

主权项

1. A method comprising:
receiving a plurality of three-dimensional (3D) object data models, the plurality of 3D object data models associated with a first category of objects, wherein respective 3D object data models of the plurality of 3D object data models comprise one or more annotated parts; receiving a candidate 3D object data model that is associated with the first category of objects, the candidate 3D object data model comprising one or more un-annotated parts; and based on a shape and an appearance of the one or more annotated parts and spatial-layout information that defines how the one or more annotated parts are spatially arranged, applying, by a computing device, a first annotation of a respective part of a respective 3D object data model to a corresponding un-annotated part of the candidate 3D object data model, wherein the spatial-layout information comprises information about at least one normalized distance between parts of the one or more annotated parts and about at least one relative orientation between parts of the one or more annotated parts, and wherein applying the first annotation comprises:
determining one or more first shape-context descriptors for the respective 3D object data model, wherein the one or more first shape-context descriptors comprise a first shape-context descriptor associated with a first point of the respective 3D object data model, wherein the first shape-context descriptor is configured to capture a distribution of points in the respective 3D object data model relative to the first point;determining one or more second shape-context descriptors for the candidate 3D object data model, wherein the one or more second shape-context descriptors comprise a second shape-context descriptor associated with a second point of the candidate 3D object data model, wherein the second shape-context descriptor is configured to capture a distribution of points in the candidate 3D object data model relative to the second point;determining an aligning transformation for aligning the 3D object data model with the candidate 3D object data model, wherein the aligning transformation is based on correspondences between the one or more first shape-context descriptors and the one or more second shape-context descriptors; andapplying the first annotation to the corresponding un-annotated part of the candidate 3D object data model based on the aligning transformation.