Asst. Prof. Dr. Ayşegül Dündar’s project on “Semantically conditioned multi-modal image inpainting” received TÜBİTAK 3501 support.
Image inpainting, the task of filling missing pixels in the most realistic manner, is a difficult task requiring a high level of understanding of the scene. A successful image inpainting algorithm provides an important image editing tool that can be used by daily users as well as architects and designers. Recently, deep learning methods have also dominated this task as it happened in other computer vision tasks. Even though much progress has achieved in this domain, results are far from perfect. In this project, a multimodal representation for semantically conditioned image inpainting algorithm will be designed to synthesize diverse results and provide controllobility to the user of the inpainted area.