High-resolution point cloud reconstruction from a single image by redescription

Point cloud is widely used because of its vectorized and compact information, but the unorder and discrete characteristics reduce the accuracy of point cloud reconstruction from images. In this paper, we propose a novel method to reconstruct high-resolution object point cloud by image redescription and point cloud upsampling. We first combine reconstruction and upsampling networks to generate high-resolution point cloud and achieve joint optimization through phased training. Then we present an image redescription mechanism to achieve the bidirectional correlation and enhance the semantic consistency between images and point clouds. The experiments on the ShapeNet dataset demonstrate the superiority of the proposed method over the state-of-the-art methods.