Data-driven Structure-adaptive Editing of Man-made Objects
Description3D modeling is a fundamental problem in 3D computer graphics. Although many 3Dmodeling techniques have been proposed, 3D modeling is still one of the bottlenecks formany applications such as virtual reality and 3D printing. In this project, we focus onthe problem of interactive shape editing, which is a popular modeling paradigm and aimsto generate new variations of a given shape via a moderate amount of user interactions.The recent efforts have been mainly put to the design of editing frameworks for manmadeobjects, which are highly demanded for 3D printing and the digitization of our realworld. Most of the existing man-made shape editing techniques aim for the creation ofgeometric variations of a shape, typically through structure-preserving shapedeformation. These techniques thus do not support either implicit or explicit explorationof structural variations during editing. The creation of new shape structures has beenstudied recently in context of synthesis from shape collections but is not directlyapplicable for editing a given shape.We observe that desired editing effects sometimes might be difficult to achieve withoutchanging the shape structure of a shape. This motivates us to design a shape-adaptiveediting framework which automatically decides whether to keep or change the shapestructure to reflect user inputs. Our key idea is that possible geometric variations can bemodeled by a group of shapes with similar structures, and possible structural variationscan be determined by performing inter-group shape analysis. This leads to a data-drivenapproach with three key problems involved: How to cluster various shapes in thesame category into structure groups? How to learn group-sensitive priors? How toadaptively edit shapes with the assistance of the learned group-sensitive priors?Our technique will automatically bring geometric and/or structural changes to a shapebeing edited dependent on user inputs. We believe that by allowing automaticadjustment of shape structures our technique will produce structurally more reasonableediting results. We also plan to explore various user interfaces to support a more explicitway for editing shape structures, making it possible to produce both geometric variationsand structure variations when editing a single shape. With the growing accessibility ofman-made objects, we expect that benefited from the rich knowledge embedded in thedatabase, our structure-adaptive editing approach will enable both professional andnovice users to easily produce new shapes, which are readily used by variousapplications.
|Effective start/end date||1/01/17 → 24/06/21|
- 3D Modeling , Interactive Shape Editing , Structures , Data-driven Approaches , Man-made Objects