Brief

Researchers have introduced VecSet-Edit, a novel pipeline for editing 3D meshes from a single image, addressing limitations in existing methods. This approach utilizes the VecSet Large Reconstruction Model (LRM) by analyzing its spatial token properties to identify distinct geometric regions. The system employs strategies like Mask-guided Token Seeding and Attention-aligned Token Gating for precise localization and a Drift-aware Token Pruning mechanism to handle geometric outliers, ensuring high-fidelity mesh editing. AI

Researchers have introduced 3D-ReGen, a novel framework for regenerating 3D objects from 2D images and existing 3D models. Unlike one-shot generators, 3D-ReGen is conditioned on an initial 3D shape, enabling tasks such as enhancement, reconstruction, and editing. The system utilizes a new conditioning mechanism called VecSet for detailed geometric updates and learns regeneration priors through self-supervised learning on large 3D datasets without requiring additional annotations. Evaluations show 3D-ReGen achieves state-of-the-art performance in controllable 3D generation. AI