3D-Aware Object Editing for Indoor Scenes

High-quality indoor scene understanding faces inherent dual challenges in complex geometry recovery and photorealistic image restoration, particularly when the method is used for joint object removal and 3D reconstruction. We present a unified 3D-aware perception framework that integrates three synergistic components: an instance segmentation network for precise object mask extraction, a diffusion-based inpainting module for context-aware texture synthesis, and a monocular depth estimation branch for geometrically consistent scene representation. By jointly optimizing RGB imagery, predicted depth maps, and camera parameters, our method achieves simultaneous high-fidelity object removal and metrically accurate 3D reconstruction. The depth estimation provides explicit geometric constraints during inpainting, while segmentation masks guide object-aware depth refinement, ensuring spatial coherence between restored textures and scene geometry. Extensive experiments on multiple benchmark datasets demonstrate that our approach consistently outperforms existing methods in both visual realism and reconstruction accuracy, establishing a solid foundation for the development of spatially-aware intelligent home systems.