Image-based 3D object reconstruction: state-of-the-art and trends in the deep learning era

Abstract

Image-based 3D reconstruction is a very challenging problem in computer vision and deep learning. Since 2015 image-based 3D reconstruction using a convolution neural network has attracted and demonstrated impressive performance. We focus on the work that uses deep-learning techniques to reconstruct the 3D shape of generic objects from single or multiple RGB images. However, unlike 2D images, 3D cannot be represented in its canonical form to make it computationally lean and memory-efficient. This paper proposes Grid/voxel-based 3D object reconstruction from a single 2D image for better accuracy, using the Autoencoders (AE) model. The encoder part of the model is used to learn suitable compressed domain representation from a single 2D image, and a decoder generates a corresponding 3D object. We provide a comprehensive, structured review of the recent advanced 3D objects reconstruction using deep-learning techniques.