SinBlowUp: Outpainting from Learning a Single Image

Abstract

We propose an outpainting method that generates the outer region of an image only by learning the input image itself, without a training stage in advance. By modifying the previous SinGAN, we generate various outpainting images of the same style as the input image. We call the proposed method SinBlowUp and we showed that this can be used for both a photo-image and an art painting with various experimental results. Proposed method can be used to create exhibition contents for daily display in home appliances such as frame-type TVs or VR devices.