Method and device for retrieving corresponding three-dimensional model based on two-dimensional image

Abstract

The invention discloses a method and device for retrieving a corresponding three-dimensional model based on a two-dimensional image. The method comprises the following steps: acquiring a to-be-retrieved two-dimensional image and retrieving two-dimensional image features; obtaining a database, wherein the database comprises a plurality of two-dimensional pictures and a plurality of three-dimensional models; obtaining features of each two-dimensional picture and features of each three-dimensional model; establishing a relation graph; according to the relation graph, constructing a plurality of fully connected graphs through a multi-head attention method; generating an embedded graph according to the full connection graph; fusing the plurality of embedded graphs to generate a unique embedded graph; and performing similarity comparison on the to-be-retrieved two-dimensional image features and each node representing the three-dimensional model in the unique embedded graph to obtain the node with the highest similarity, and outputting the three-dimensional model corresponding to the node with the highest similarity. According to the method, the correlation between the 2D image and the 3D model is fully considered, and the difference between two different information retrieval modes is eliminated.

Description

technical field [0001] The invention relates to the technical field of three-dimensional model retrieval, in particular to a method for retrieving a corresponding three-dimensional model based on a two-dimensional image and a device for retrieving a corresponding three-dimensional model based on a two-dimensional image. Background technique [0002] With the development of deep learning methods, 3D model retrieval has become a hot topic in the field of computer vision due to its compatibility with various applications such as computer-aided design, medical diagnosis, bioinformatics, 3D visual saliency, 3D printing, digital entertainment, etc.) has received extensive attention from researchers. [0003] The surge in the number of 3D models has created a huge demand for efficient 3D model retrieval algorithms. Meanwhile, with the development of smart phones, people can easily obtain 2D images. Therefore, given a 2D image, retrieving a 3D model is becoming a more convenient w...

AI Technical Summary

Problems solved by technology

[0008] 2) While ensuring the accuracy of classification and retrieval, the computational time and space complexity must also be considered. At present, many 3D model retrieval tasks simply increase the depth of the training network and lack more effective algorithms.

Images