Automatic Generative Furniture Arrangement Method Based On Graph Neural Network

Automatic furniture placement tasks have a wide range of application scenarios in the fields of indoor home improvement design,virtual reality,computer vision and other fields that require digital indoor scenes.The traditional method uses sequential iteration or energy optimization to arrange furniture after pre-defining rules such as the location,function,and rationality of the furniture.However,the definition of rules and the process of optimizing iterations will cause a lot of manpower and time consumption.With the emergence of large-scale indoor scene datasets,people have begun to explore the use of deep learning for furniture placement.Therefore,a neural network model that can complete the furniture placement task under the given furniture and room structure conditions,as well as a data structure capable of training neural networks and expressing indoor scenes are needed.Different from the existing hierarchical structure and other indoor scene expression methods that can only transmit information in one direction,this paper proposes to use the graph convolutional neural network that can transmit information in both directions to encode and decode the indoor scene layout.At the same time,in order to realize the generation of diverse placement methods under given conditions,this paper adopts the structure of conditional variational autoencoder to fit the real scene layout information into the hidden space of the standard Gaussian distribution,and the scene prior of Gaussian distribution sampling is used to guide the subsequent placement of scattered,unstructured furniture,and finally generate the position and orientation of the furniture layout information for each furniture.This paper proposes a new way of using graph convolutional neural network and conditional variational autoencoder to encode and generate scenes.And the visual results and quantitative indicators of subsequent experiments all prove the effecti veness of the algorithm in this paper.