Research on 3D Object Reconstruction Based on Single-view RGB Image

With the development of immersive virtual reality technology, how to quickly and conveniently perform three-dimensional (3D) reconstruction of objects has become a hot research topic. In this paper, we propose a method to reconstruct 3D objects from single view RGB images. The depth SDF representation of 3D shape is predicted from the input image, and the shape is expressed in a compact way, which can predict the shape with complex topology and retain the shape details. In view of the lack of prior use of the shape of similar objects in the prior art, which may generate incomplete and unrealistic shapes, an implicit template shape is constructed in this paper, and a coordinate transformation field of the implicit template shape is proposed to be estimated from the input RGB image, and the SDF of the implicit template is transformed into the SDF of the estimated shape. The shape prior contained in the template shape is used, which improves the robustness of the shape estimation.