SPGen: Spherical Projection as Consistent and Flexible Representation for Single Image 3D Shape Generation

2025-09-16

AI Summary

Existing single-view 3D generation methods rely on multi-view diffusion priors, suffering from view inconsistency and struggling to model complex internal structures and non-trivial topologies. To address this, we propose Spherical Projection (SP), a novel 3D representation that maps shape geometry onto an enclosing sphere and unfolds it into a multi-layer 2D structure, enabling consistent and flexible single-image-driven reconstruction. By incorporating a single-view injection mechanism into the SP mapping, we eliminate inter-view contradictions and support generation of watertight/open surfaces as well as nested internal structures. Joint fine-tuning of a 2D geometric encoder and 2D diffusion priors over the SP representation ensures computational efficiency while significantly improving geometric fidelity. Experiments demonstrate that our method surpasses current state-of-the-art approaches under limited computational resources, achieving simultaneous advances in reconstruction consistency, topological expressiveness, and inference efficiency.

Technology Category

• Computer Vision: 3D Computer Vision
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Existing single-view 3D generative models typically adopt multiview diffusion priors to reconstruct object surfaces, yet they remain prone to inter-view inconsistencies and are unable to faithfully represent complex internal structure or nontrivial topologies. In particular, we encode geometry information by projecting it onto a bounding sphere and unwrapping it into a compact and structural multi-layer 2D Spherical Projection (SP) representation. Operating solely in the image domain, SPGen offers three key advantages simultaneously: (1) Consistency. The injective SP mapping encodes surface geometry with a single viewpoint which naturally eliminates view inconsistency and ambiguity; (2) Flexibility. Multi-layer SP maps represent nested internal structures and support direct lifting to watertight or open 3D surfaces; (3) Efficiency. The image-domain formulation allows the direct inheritance of powerful 2D diffusion priors and enables efficient finetuning with limited computational resources. Extensive experiments demonstrate that SPGen significantly outperforms existing baselines in geometric quality and computational efficiency.

Problem

• Addresses inter-view inconsistencies in single-view 3D generation
• Represents complex internal structures and nontrivial topologies
• Enables efficient 3D shape generation using 2D diffusion priors

Innovation

• Spherical projection for 3D shape representation
• Multi-layer 2D mapping eliminates view inconsistency
• Image-domain formulation enables efficient diffusion priors