The article discusses ‘RealFill’, a novel generative approach developed by Google Research and Cornell University for image completion. It uses a few reference images to fill in missing regions of an image with the content that should have been there. It can handle varying viewpoints, lighting conditions, camera apertures, or image styles. RealFill outperforms existing approaches by a large margin in completing a target image with content that is faithful to the original scene.
Publication date: 28 Sep 2023
Project Page: https://realfill.github.io
Paper: https://arxiv.org/pdf/2309.16668
