Research on Single-View 3D Reconstruction Algorithm for Grouped Convolutional Coding
In order to further improve the accuracy of 3D reconstruction of single view image,an improved single view 3D reconstruction network was proposed by studying the current advanced algo-rithms.By improving the feature extraction network,the encoder of the network could obtain more abundant,complete and deep two-dimensional features.The attention mechanism was introduced into the network architecture of the refiner to further refine the 3D features and generate a more refined 3D voxel model.In addition,threshold adjustment module was added in the network to make up for the differences between different kinds of images to achieve better reconstruction effect.The experimental results show that the overall IoU value of 3D reconstruction on the public data set ShapeNet reaches 0.675,and the network achieves better results in single image reconstruction.
NETL
NSTL
万方数据
