Relies on the deep-image-prior research by Dmitry Ulanov et.al.
This repository uses the method of image inpainting using deep image prior primarily used for restoring cave paintings images. it uses a CNN in conjunction with a GAN to identify pixels in the inpainting region which actually represent unscathed parts of the painting. Has been found to be capable of restoring depicted ornaments and facial features including facial expressions of people and deities in multiple paintings in Ajanta Caves, Mumbai. Has achieved results close to ground-truth (tested on paintings whose older images were available)