Abstract
Retrieving missing data from a given piece of information has always been a challenging task. “Inpainting” is one of the techniques which helps us to solve such problems. It basically fills in the missing data such that the filled data is contextually relevant to its surroundings. This project aims in automating Inpainting process with minimal or no human intervention. The idea of Generative Models is used here, which understands the hidden structure of the given data set of particular distribution in order to generate new samples from the same distribution. Implementation of the project is based on Raymond Yeh et al. methodology for Inpainting which is analogous to how human brain completes a missing image. The goal here is not to reconstruct an image using generative models, but instead to use them to fill in the missing holes with more realistic content. The project currently focuses only on the regular (center) missing regions in an image. An attempt to experiment with Hamiltonian Monte Carlo (HMC) optimization algorithm apart from regular Adam optimization is made in this project. There are many practical applications where Inpainting can be used. One such application would be in the field of criminology. Systems like this can be used to generate images of culprits when the information on their appearance is incomplete.