Object Removal and Inpainting from Image using Combined GANs

Abstract

As recent research on deep learning methods has been actively conducted, a number of deep learning methods have been proposed. In this paper, we propose a method of removing the desired object from an image using generative adversarial networks(GANs) structure. We composed the network in which two GANs are fused. The first GAN erases the target object from the input image, and the second GAN generates an image that fills the empty space with the background. Through this network, we can erase the desired object from the input image and get an image with the erased part filled with the background without any object detection method. We show that the removal of people and vehicles from images of roads using the CityScapes Dataset.