An Interior Intelligent Design System Based on GAN and 3D Scene Automatic Generation Technology

Abstract

To address the inefficiencies, high costs, and limitations in spatial rationality and automation prevalent in traditional and existing intelligent interior design approaches, this study proposes and implements an end-to-end intelligent interior design system. The system integrates a deep convolutional Generative Adversarial Network (GAN) for automated floor plan layout generation with an automatic 3D scene generation module, facilitating a seamless transition from 2D layouts to immersive 3D visualizations. The proposed GAN model incorporates class-aware convolutional layers and a spatial constraint loss function to enhance layout rationality and diversity. Experiments demonstrate that the proposed method achieves high layout accuracy (96.78%) and low furniture overlap (7.28%) on public datasets, while maintaining robust 3D reconstruction accuracy (84.36%) and precise model alignment. System performance tests show real-time rendering capability under concurrent user loads. This research provides an efficient automated solution that enhances design productivity, supports visual decision-making, and offers practical value for project management in interior design applications.