VecSet-Edit Enables Single-Image 3D Mesh Editing via Token Localization

VecSet-Edit uses a VecSet Large Reconstruction Model backbone to edit 3D meshes directly from a single 2D image, bypassing the resolution limits of voxel-based approaches. Mask-guided Token Seeding and Attention-aligned Token Gating localize target geometry regions using only 2D image conditions.