A Novel Method for Image Inpainting Using CGAN

Image Inpainting can be defined as the task of filling out blank spaces or holes that may appear on images. It has been mostly used for restoring old and damaged photos. Traditionally methods like image segmentation and auto encoders were used to obtain results for this problem. Also, the old traditional methods had to have datasets where the developer would manually pick which areas for the model to train so that it can paint those missing regions correctly. Newer deep learning-based techniques, which may also be carried out unsupervised, have demonstrated remarkable performance in producing visually realistic and refined contents for the missing regions in free-form image inpainting challenges. Although several existing solutions concentrate on including more inputs to address this issue, it is difficult to directly apply cutting-edge inpainting techniques to image extension since they frequently produce blurry or repeating pixels with erratic semantics. The discriminator of a generative adversarial network (GAN) is subjected to semantic conditioning, and the findings on image extension with coherent semantics and aesthetically pleasant colors and textures are strong. We also present encouraging findings for long extensions.