The Attribute Driven Automatic Generation Method For 3D Virtual Exhibition Layouts

Scene layout modeling and generation is an important research topic in the field of artificial intelligence and computer graphics.At present,people have studied more about the generation of home and office scenes,and rarely pay attention to the automatic layout generation of virtual exhibitions.Therefore,this paper conducts the first attempt to adopt a data-driven method to automatically generate the virtual exhibition layouts.Specifically,we establish a metadata description of the virtual exhibition layouts,and collect the corresponding data set.Then,we propose an attribute-driven automatic generation method for 3D virtual exhibition layouts.In a specific space-time range,the generation of the virtual exhibition layouts need to use artistic design knowledge to create the space or plane carefully,to determine the best spatial relationship of the objects in the scene.However,virtual exhibitions are mainly created through real scene scan or interactive design now,which are inefficient and lack universality.In order to improve the efficiency of visual exhibition design and generate more diverse layouts,this paper proposes the attribute-driven automatic generation method for 3D virtual exhibition layouts:First,we propose a metadata description for the virtual exhibition layout,which views the virtual gallery as three parts:room,paintings and decorations,and uses corresponding data attributes to describe every part,to adaptively generate exhibition halls of arbitrary shape and scale.Then,we develop a platform for the labeling of 3D visual exhibition layouts,to label and collect virtual exhibition layout data sets that can be used for data-driven modeling.Finally,by considering the basic design concept of the virtual exhibitions,we propose an attributedriven automatic generation method for 3D virtual exhibition layouts.Specifically,this method regards the exhibition layout process as an iterative painting placement process:first trains a saliency distribution prediction model to predict the saliency of each position on the contour of the exhibition hall;then,considering the existing layout,the contour saliency,and the importance and category of the paintings,trains a location distribution prediction model to predict the location of each painting,and then generates the entire layout iteratively.Extensive comparative experiments and user survey results show that:By effectively considering the painting attributes,our method can generate diverse virtual exhibition layouts,which conform to the designer’s basic design rules.Compared with other methods,it has stronger universality and higher design efficiency.