METHOD AND DEVICE FOR CLASSIFYING AND SEARCHING FOR A 3D MODEL ON BASIS OF DEEP ATTENTION

The application discloses a method and a device of classifying and/or retrieving a three-dimensional model on basis of depth attention. The method of classifying and/or retrieving the three-dimensional model on basis of depth attention comprises the steps of: acquiring a three-dimensional model to be retrieved; mapping to generate a group of two-dimensional views according to the three-dimensional model to be retrieved, wherein the group of two-dimensional views comprises at least two two-dimensional views; acquiring features of each two-dimensional view; fusing the features of all two-dimensional views via a depth attention network to form a fused feature; and retrieving or classifying the three-dimensional model to be retrieved according to the fused feature. By introducing a self-attention structure, correlation information among multiple views can be fully considered from a global angle to mine hidden information and reduce information redundancy.