3D-ReGen: A Unified 3D Geometry Regeneration Framework

2026-04-30

AI Summary

This work addresses the limitations of existing single-pass 3D generation methods in controllability and detail consistency by proposing a unified 3D geometric regeneration framework. Conditioned on an initial 3D shape, the framework leverages self-supervised learning from general 3D data to capture regeneration priors, enabling tasks such as reconstruction, enhancement, and editing. Its core innovation lies in a VecSet-based fine-grained conditioning mechanism that precisely updates input geometry details, combined with a self-supervised pretraining strategy that requires no additional annotations, thereby establishing a general and controllable regeneration capability. Experiments demonstrate that the method achieves state-of-the-art performance across multiple controllable 3D generation tasks, significantly outperforming existing approaches—particularly in geometric consistency and fine-grained quality.

Abstract

We consider the problem of regenerating 3D objects from 2D images and initial 3D shapes. Most 3D generators operate in a one-shot fashion, converting text or images to a 3D object with limited controllability. We introduce instead 3D-ReGen, a 3D regenerator that is conditioned on an initial 3D shape. This conceptually simple formulation allows us to support numerous useful tasks, including 3D enhancement, reconstruction, and editing. 3D-ReGen uses a new conditioning mechanism based on VecSet, which allows the regenerator to update or improve the input geometry with consistent fine-grained details. 3D-ReGen learns a widely applicable regeneration prior from off-the-shelf 3D datasets via self-supervised pretext tasks and augmentations, without additional annotations. We evaluate both the geometric consistency and fine-grained quality of 3D-ReGen, achieving state-of-the-art performance in controllable 3D generation across several tasks.

Problem

• 3D regeneration
• 3D generation
• geometry consistency
• fine-grained details
• controllable 3D

Innovation

• 3D regeneration
• VecSet conditioning
• self-supervised learning
• controllable 3D generation
• geometry enhancement