Indoor Home Scene Generation Algorithm Based On GAN

The computer vision field is getting more and more attention.Among them,scene generation problems involve various aspects such as Games' map modeling,home decoration,and urban object layout.The broad market prospects have prompted more and more researchers to invest in this issue.Currently,this problem mainly includes hardware methods and software methods.hardware approach require harsh equipment and environmental,anything equipment scanned from the sample will completely determine the generated sample.The diversity of scene generation is greatly limited.This method is becoming more and more mainstream by selecting layout objects from the case library and then laying out them in a given space.However,there are still two major problems for different models.One is that the algorithm is time consuming and cannot solve the layout problem of too many objects and complex layout spaces.The other is that the layout is not interactive and lacks diversity.The situation of a scene layout completely mimics the data set and cannot be added to people's subjective intentions or suggestions for the layout.In view of the above problems,we start to study the high-performance interactive indoor furniture layout algorithm based on the generation of indoor scenes.The main contributions are as follows:We propose a method based on functional area division and furniture filling to solve the layout problem with too many layout objects in complex living room environments.According to the function,we suppose each kind of furniture may be laid out in one or several functional areas,for example,a sofa may be located in the meeting area of a living room and a bed may be located in the sleeping area of a bedroom,etc..Our automatic layout method divides an empty room region into several functional areas by using conditional generative adversarial networks(CGAN).We expound the learning proceeding for functional area division,including the objective function construct and training process.Moreover,in order to fill furniture into a specific functional area,a learning based furniture filling algorithm is proposed by training a fully connected network model for every type of functional area.Experiments show our automatic furniture layout method has its advantages in performance and effect compared with the existing methods.The most important problem that we have solved in the previous work is the huge time consumption of the algorithm in complex scenarios.And compared to traditional hardware-based device scanning methods.The effect of our algorithm is also improved in the diversity of the layout.But we hope that our scene generation is interactive.Therefore we build a TextGAN network structure to establish an association between the scenario and the interactive code.We strive to generate scenes that are interactive and diverse.This requires that our data set must be rich in content and contain interactive information.This poses a challenge to the training of the model.So we made data sets and adjust the training approach of the model.