LLM-driven Indoor Scene Layout Generation via Scaled Human-aligned Data Synthesis and Multi-Stage Preference Optimization

2025-06-09

AI Summary

Automatic indoor layout generation faces challenges including spatial inconsistency, high computational cost, coarse relational modeling, and poor generalization. This paper proposes OptiScene: first, it establishes a “GPT-synthesized + human-verified” paradigm to release 17K high-quality 3D-SynthPlace layouts; second, it introduces a two-stage preference optimization framework—the first to apply multi-round Direct Preference Optimization (DPO) to layout generation—integrating supervised fine-tuning (SFT) of open-source LLMs with high-order spatial description modeling and conditional object placement. Experiments demonstrate that OptiScene significantly outperforms prompt-driven and diffusion-based baselines in layout plausibility, generation success rate, and downstream tasks (e.g., interactive editing, robot navigation), markedly improving spatial consistency and alignment with human design preferences.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Planning, Routing, and Scheduling: Planning with Language Models
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Automatic indoor layout generation has attracted increasing attention due to its potential in interior design, virtual environment construction, and embodied AI. Existing methods fall into two categories: prompt-driven approaches that leverage proprietary LLM services (e.g., GPT APIs) and learning-based methods trained on layout data upon diffusion-based models. Prompt-driven methods often suffer from spatial inconsistency and high computational costs, while learning-based methods are typically constrained by coarse relational graphs and limited datasets, restricting their generalization to diverse room categories. In this paper, we revisit LLM-based indoor layout generation and present 3D-SynthPlace, a large-scale dataset that combines synthetic layouts generated via a 'GPT synthesize, Human inspect' pipeline, upgraded from the 3D-Front dataset. 3D-SynthPlace contains nearly 17,000 scenes, covering four common room types -- bedroom, living room, kitchen, and bathroom -- enriched with diverse objects and high-level spatial annotations. We further introduce OptiScene, a strong open-source LLM optimized for indoor layout generation, fine-tuned based on our 3D-SynthPlace dataset through our two-stage training. For the warum-up stage I, we adopt supervised fine-tuning (SFT), which is taught to first generate high-level spatial descriptions then conditionally predict concrete object placements. For the reinforcing stage II, to better align the generated layouts with human design preferences, we apply multi-turn direct preference optimization (DPO), which significantly improving layout quality and generation success rates. Extensive experiments demonstrate that OptiScene outperforms traditional prompt-driven and learning-based baselines. Moreover, OptiScene shows promising potential in interactive tasks such as scene editing and robot navigation.

Problem

• Addressing spatial inconsistency in prompt-driven layout generation
• Overcoming dataset limitations in learning-based layout methods
• Improving generalization across diverse room categories

Innovation

• LLM-driven layout generation with human-aligned data
• Two-stage training: SFT and multi-turn DPO
• Large-scale dataset 3D-SynthPlace for diverse rooms