Research on 3D Model Set Matching Based on Unsupervised Cyclic Consistency Network

To solve the problem of correspondence calculation of non-rigid 3D model set, an Unsupervised Cycle Consistency Network (UCCN) is proposed to calculate the dense correspondence of non-isometric 3D model set. First, the parameter prediction network is used to predict the deformation parameters between the two models, and then the source model is transformed into the target model through the deformation network, Finally, the cycle consistency constraint is introduced to achieve the consistency of the mapping relationship between multiple models in the model set, and the corresponding relationship between models in the model set is calculated. The experimental results show that the algorithm can effectively improve the accuracy of correspondence calculation of 3D model set. Compared with the traditional method, it realizes the mechanism of unsupervised deep learning, and does not depend on the influence of intermediate template on matching accuracy.