An Improved GAIN-Based Model for Image Imputation

Abstract

Missing data represent a main issue having a considerable impact not only on the inference quality but also on the prediction relevance. A problem that researchers have constantly tried to solve, ranging from Statistics to Deep Learning through Machine Learning. Recently, a growing interest is being given to Deep Learning models and particularly to Generative Adversarial Networks (GAN) given their interesting performance and their adaptability to various scenarios. In this paper, we propose an improvement of Generative Adversarial Imputation Nets (GAIN) which is one of the main GAN-based models having proved its worth in image imputation. The results found confirm the proposed improvement efficiency.