Bilkent University
Department of Computer Engineering


Image Feature Manipulation and Feature Space Interpretation in Generative Models


Yusuf Dalva
MS Student
(Supervisor: Asst. Prof. Ercüment Çiçek)
Computer Engineering Department
Bilkent University

"Image manipulation and translation studies have made significant developments in generative models in both supervised and semi-supervised settings. Since the emergence of style-based manipulation with [1], image generation with specific attributes started to give promising results. Until today, several studies attempted to uncover the learning process in generative models to understand attribute specific learning. By starting from a semi-supervised setting where the target feature in manipulation is not predefined by model definition, newer studies are able to perform it in a supervised way [2]. Understanding this type of manipulation improves with new datasets [3] and new studies. The primary purpose of this presentation is to explain the state-of-art methods in the image feature manipulation area. In this regard, the audience will be informed about both the revolutionary studies that changed the research direction and state-of-art methods that perform best in today’s conditions. Also, this presentation aims to explain the approaches used to interpret the feature space learned by the models and how this information may contribute to future studies.detected keypoints.


DATE: 08 November 2021, Monday @ 16:10 Zoom