刘明华.pptx

1、PartField,Learning 3D Feature Fields for Part Segmentation and BeyondMinghua Liu04/11/2025,3D AIGC Ready to Use?,much more accurate geometry and texture than before,AIGC vs.Artist Created,holistic meshdifficult to edit,We need hierarchical parts!,w/hierarchical groupingeasy to edit,Where Else do We

2、Need Parts?,physical simulation,UV mapping,animation&rigging,What Parts do We Need?,Open-worldMulti-granularity/hierarchical partsGeneral concept of parts,not necessarily semantic partsFast and robust inference,PartField,We train a feedforward model that takes a point-sampled 3D shape as input and p

3、redicts a feature field represented by a triplane.These features model the general concept of 3D parts and can be clustered to generate parts at various scales.,Speed and Accuracy,Hierarchical Part Decomposition,Various Input Modalities and Styles,Emerged Cross-Shape Consistency,Co-Segmentation,Shape Correspondence,Follow-Up:Part-Aware UV Unwrapping,XATLAS(184 charts),Blender SmartUV(691 charts),Ours(45 charts),PartField,Learning 3D Feature Fields for Part Segmentation and BeyondThank you!,