Point cloud semantic segmentation network model of voxelization network based on deep residual error and attention discarding

The invention relates to the technical field of computer three-dimensional visual processing, in particular to a point cloud semantic segmentation network model of a voxelization network based on deep residual error and attention discarding, which comprises a Deep Inversion Residual module and a Drop Attention module, and is characterized in that the Deep Inversion Residual module is utilized to mine voxel deep information so as to make up for the loss of point cloud information loss during voxelization, and then the Drop Attention module is utilized to extract the point cloud semantic segmentation network model of the voxelization network based on deep residual error and attention discarding. The receptive field range of each voxel for global information is expanded; according to the method, deep information of voxels is mined by utilizing Deep Inverse Residual, so that loss caused by point cloud information loss in a voxelization process is compensated; and enabling the local voxel region to obtain global i