StyleForge: Indoor Furniture Styling by Counterfactual Reasoning in a Hypergraph Field

2026-08-03

AI Summary

Existing approaches to indoor furniture arrangement often suffer from style inconsistencies in shape, material, and color due to either independent selection or reliance on static, local relationships. To address this, this work proposes StyleForge, a framework that models high-order furniture dependencies through a dynamic hypergraph-based style field. It leverages a frozen multimodal large language model to extract stylistic priors and integrates counterfactual preference learning with a Mahalanobis distance-based energy function to enable context-aware, scene-level style harmonization. The method supports test-time adaptation by updating only room-specific candidate distributions to resolve cross-slot conflicts. Evaluated on the 3D-FRONT dataset, StyleForge significantly outperforms existing object-level and scene-level retrieval baselines, achieving state-of-the-art performance in both furniture retrieval accuracy and overall stylistic coherence.

Abstract

Fixed-layout indoor furniture styling requires selecting assets that form a coherent room without changing the prescribed furniture categories, positions, orientations, or scales. Existing approaches typically retrieve each asset independently or rely on static local relations, making them prone to shape, material, and color conflicts after scene composition. We introduce StyleForge, a scene-level structured selection framework built on a dynamic hypergraph style field. A frozen multimodal large language model extracts structured style priors from an open-ended style request and the fixed layout, while StyleForge maintains a learnable candidate distribution for each furniture slot. Conditioned on the target style, the dynamic hypergraph style field adaptively activates and weights layout-induced hyperedges to capture higher-order dependencies among furniture. Counterfactual style preference learning then treats each candidate as a local substitution in the current style field and evaluates its contextual compatibility using Mahalanobis energies. Training alternates between optimizing the style field and the candidate logits. At inference, the model remains frozen and test-time training updates only room-specific candidate logits, progressively correcting cross-slot style conflicts as the global scene context evolves. Experiments on 3D-FRONT demonstrate state-of-the-art furniture retrieval and scene-level style coherence, producing more coherent fixed-layout furniture arrangements than object- and scene-level retrieval baselines.

Problem

• indoor furniture styling
• fixed-layout
• style coherence
• scene composition
• furniture arrangement

Innovation

• counterfactual reasoning
• hypergraph field
• style coherence
• structured selection
• test-time training