Generalized efficient Gaussian representation learning method and device, equipment and storage medium

Abstract

The invention relates to the technical field of three-dimensional reconstruction, in particular to a generalizable efficient Gaussian representation learning method and device, equipment and a storage medium. The method comprises the steps that a plurality of input images are acquired, a Gaussian graph structure is established according to the input images, and the Gaussian graph structure is used for indicating the mutual relation between three-dimensional Gaussian representations corresponding to the input images; and processing the Gaussian map structure through a preset Gaussian map neural network to obtain a target Gaussian representation of the three-dimensional scene, the Gaussian map neural network being used for fusing the three-dimensional Gaussian representations corresponding to the plurality of input images, and the target Gaussian representation being used for reconstructing the three-dimensional scene. According to the method provided by the embodiment of the invention, the problem of Gaussian redundancy caused by a current three-dimensional reconstruction method is solved by modeling the relationship between the three-dimensional Gaussian representations of different visual angles, and the precision, efficiency and generalization ability of three-dimensional scene reconstruction are improved.