Look Both Ways Before You Cross: Lifting Cross Fields From 2D Visual Priors

Researchers introduced CrossLift, a mesh-processing method that lifts 2D visual priors into 3D cross fields.

It extracts directional cues from text-to-image generated images, back-projects them onto meshes, and interpolates them across views and surfaces.

The resulting cross fields help guide quad meshing and related design tasks with better alignment to visual features.

The approach is modular and supports interactive workflows on complex shapes, improving semantic consistency in mesh generation.