Pyramid Transform-based point cloud reconstruction method, device and equipment, and medium

Abstract

The invention discloses a pyramid Transform-based point cloud reconstruction method, device and equipment, and a medium, and belongs to the field of computer vision. The method comprises the steps that: after point feature data corresponding to each view are obtained, the point feature data are input into a two-channel pyramid network, and point features input in each layer pass through three modules of sampling, a graph neural network and Transformer; when the point feature data passes through the first channel, the updated output of each layer of attention mechanism is input into the next layer, and when the point feature data passes through the second channel, the updated output of each layer of attention mechanism and the updated output of the attention mechanism of the corresponding layer of the first channel located in the same layer as the next layer are spliced and then input into the next layer; and the updated output of the attention mechanism of the bottommost layer of the second channel is converted into a point cloud coordinate to obtain a point cloud result of each view, and an intersection of the point cloud results of each view is taken to represent a target object after point cloud reconstruction. Therefore, the three-dimensional reconstruction precision of the object is higher.

Description

technical field [0001] The invention belongs to the field of computer vision, and more specifically relates to a method, device, equipment and medium for point cloud reconstruction based on a pyramid Transformer. Background technique [0002] The multi-view stereoscopic 3D reconstruction technology reconstructs a 3D object by using multiple input pictures of different perspectives. This technology studies how to obtain the 3D information of the object in space through the 2D information of the picture, and finally obtain the 3D model of the object in space. The traditional 3D reconstruction technology inputs multiple pictures with different perspectives, estimates the camera parameters corresponding to each picture, and then reprojects the objects on different pixels in the pictures to the 3D space, thereby reconstructing the 3D structure of the object. However, the traditional method requires more input images, and the reconstruction effect on reflective objects and objects...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the defects and improvement needs of the prior art, the present invention provides a point cloud reconstruction method, device, equipment and medium based on a pyramid Transformer, aiming to solve the technical problem of low reconstruction accuracy of object surface details in 3D reconstruction

Images