Intelligent 3D Home Decoration Generation

Artificial intelligence has been applied in various fields, but an end-to-end customizable smart home decoration generation technique has not been found. This paper presents a multi-stage framework for generating 3D home decoration based on deep learning, aiming at the issues of insufficient customizability and 3D visualization. Firstly, YOLO is used to locate the scale region and parse the scale information; secondly, two U-Net networks, each with a shared encoder and dual decoder, are constructed to segment the furniture and rooms respectively; then, a panorama map is generated using ControlNet; finally, the effect of the generated home decoration can be visualized by a 3D viewing tool based on the sphere mapping algorithm. We conduct abundant experiments to test the results in each stage, and the visualization of results shows that our method has the potential for applications, providing an end-to-end strategy for intelligent 3D home decoration generation.