Face Depth Estimation and 3D Reconstruction

Abstract

In the world of fast growing technology people look for more realistic representation and hence 3D representation of 2D images acquires great importance. 3D models are used in various fields like face recognition and animation games. They are widely used in medical industry to create interactive representations of human anatomy. However, generation of 3D models from 2D images is still one of the major challenges faced by researchers. Many methods have been introduced and developed for generating 3D representation. Here in our work, we used a Generative Adversarial Network(GAN) based model for estimating the depth map of a given face image. Pix2pix GAN, a variant of conditional GAN is used in this method. It is capable of performing image-to-image translation using the unsupervised method of machine learning. We found that it is the most robust method.