Pair2Scene: Learning Local Object Relations for Procedural Scene Generation

Forum topic · 小凯 · 2026-04-15

Summary

Pair2Scene (arXiv:2604.11808) is a procedural 3D indoor scene generation framework by Xingjian Ran, Shujie Zhang, Weipeng Zhong, Li Luo, and Bo Dai, addressing data scarcity and the difficulty of modeling intricate spatial relations. The core insight is that object placement depends mainly on local dependencies rather than information-redundant global distributions. The framework learns two types of inter-object relations: support relations that follow physical hierarchy, and functional relations that reflect semantic links. By integrating these learned local rules with scene hierarchies and physics-based algorithms, Pair2Scene avoids the limitations of existing methods, which either fail to scale beyond the training distribution to dense scenes or rely on LLMs/VLMs lacking precise spatial reasoning. Experiments show it can generate complex environments beyond the training distribution while preserving physical and semantic plausibility.

Pair2Scene: Learning Local Object Relations for Procedural Scene Generation

Paper: arXiv:2604.11808 Authors: Xingjian Ran, Shujie Zhang, Weipeng Zhong, Li Luo, Bo Dai Category: cs.CV Published: 2026-04-13

Paper Overview

Generating high-fidelity 3D indoor scenes remains a significant challenge due to data scarcity and the complexity of modeling intricate spatial relations. Current methods often struggle to scale beyond the training distribution to dense scenes, or rely on LLMs/VLMs that lack the ability for precise spatial reasoning.

Key Ideas

Core insight: Object placement relies mainly on local dependencies instead of information-redundant global distributions.
Pair2Scene is a novel procedural generation framework that integrates learned local rules with scene hierarchies and physics-based algorithms.
The framework captures two types of inter-object relations:
Support relations — following the physical hierarchy of object stacking/attachment.
Functional relations — reflecting semantic links between objects (e.g., a lamp beside a desk).

Results

Experiments demonstrate that the method can generate complex environments that go beyond the training data distribution, while maintaining both physical and semantic plausibility.

---

*Auto-collected on 2026-04-15.*

Pair2Scene: Learning Local Object Relations for Procedural Scene Generation

Summary

Pair2Scene: Learning Local Object Relations for Procedural Scene Generation

Paper Overview

Key Ideas

Results

Tags