Research On Example Based Modeling Technology For 3D Shape Generation

As a new type of digital media,the 3D digital geometric model has been widely used in the fields of reverse engineering,industrial design,film and television enter-tainment,virtual scenes,restoration and display of cultural relics,and biomedicine.At present,there are many mature modeling technologies,such as: laser scanning sys-tem,close-up photogrammetry system,Auto CAD and Maya software.However,the specialization of modeling equipment and technology,and the complication of the modeling process have become the bottlenecks in its promotion.How to help non-professional users to quickly and efficiently create the required models has become an important research direction in the field of digital geometry.On the one hand,the pro-cess of creating a 3D model from scratch is cumbersome and highly specialized? on the other hand,3D models have been accumulated for many years,but the reuse rate is low.How to make effective use of existing models and quickly create models with similar styles or functions that meet the user's expectation—that is,case-based modeling—has become one of the concerns of current researchers.This article starts from the application requirements of case modeling,and ad-dresses the issues of modeling diversity,component selection flexibility,and modeling rationality in the process of digital geometric modeling.The in-depth and systematic research of the 3D modeling method,the innovative results obtained mainly include the following aspects:(1)A 3D model assembly method based on graph convolutional neural network is proposed.In this method,the geometric features and structural information of the components of the graph convolutional neural network are used as component feature representations to detect the context relationship inherent in the model structure,and to realize the automatic analysis and mining of the model structure to ensure the diversity of the selected components.Select components based on features and re-verify the compatibility of the components in the new model using the network in this paper,thereby ensuring the overall rationality and functional maintenance of the assembly results.(2)A 3D model assembly method based on the structure-aware model correspon-dence network is proposed.This method uses the siamese network to explore the functional unit of the model,realizes the extraction of variable-granularity functional substructures,and obtains assembly components through effective correspondence be-tween models,which solves the problem of sensitive to the pre-segmentation of the model based on the structural method? Force mechanism,which only learns the com-ponent context and does not depend on the global structure,so you can select the same structural feature components in the cross-category model set to generate a creative structure.(3)A 3D model generation method based on graph generation network is pro-posed.This method designs a graph adversarial generation network,and uses a graph encoder to learn the representation space of the model structure,thereby obtaining a high-level representation of the three-dimensional model structure information? using a decoder to generate a new model from the top of the vectors in the representation space,automatically generate models that have not been seen before in the model set?finally,use a discriminator to verify the generated results,which can not only guarantee the rationality of the decoder's generated results from a higher level,but also verify the rationality of the user's manual assembly results.