RoomRecon: High-Quality Textured Room Layout Reconstruction on Mobile Devices

A team of researchers from KAIST and other institutions has introduced RoomRecon, a novel interactive pipeline for creating high-quality, textured 3D models of indoor spaces directly on mobile devices. The system directly addresses key limitations in current 3D reconstruction tech, such as the inability to reflect minor scene changes without a full rescan, poor semantic understanding, and texturing issues that degrade visual realism. RoomRecon's core innovation is a two-phase texturing pipeline that first uses AR guidance to help users capture optimal images for texturing, then employs generative AI models to significantly enhance the final texture quality, creating better replicas of real rooms.

By strategically focusing only on permanent structural elements—walls, floors, and ceilings—the system creates a base model that is easily customizable for applications like interior design and virtual staging. This approach sidesteps the complexity of modeling transient objects like furniture. The paper, accepted to the prestigious IEEE ISMAR 2024 conference where it was a Best Paper nominee, details experiments across various indoor environments. Quantitative results and a user study demonstrate that RoomRecon outperforms existing state-of-the-art methods in both the quality of its textured outputs and its efficiency in on-device computation time, marking a significant step toward professional-grade 3D capture in the palm of your hand.