Bilkent University
Department of Computer Engineering

---

CS 590/690 SEMINAR

 

MixGAN: Dual Path StyleGAN Fusion for Diverse and Editable Inpainting

 

Mustafa Utku Aydoğdu
Master Student
(Supervisor: Asst.Prof.Ayşegül Dündar Boral)

Computer Engineering Department
Bilkent University

Abstract: This study introduces a novel framework for image inpainting by leveraging Generative Adversarial Networks (GANs), specifically StyleGAN's feature space. We tackle the challenge of seamlessly integrating erased pixels into the surrounding image context while achieving diversity and preserving editability by utilizing StyleGAN's disentangled latent space. Our approach involves two parallel generations of features: one from the encoded image and the other from sampled latent space for diversity. These parallel generations complement each other and are aware of each other's output through dual feedback mechanisms for alignment. Our approach achieves diverse inpainting and editing within a unified framework. Through experiments and comparisons with existing methods, we demonstrate the effectiveness of our approach in achieving high-quality inpaintings and significant improvements on diverse generations.

 

DATE: March 17, Monday @ 13:20 Place: EA 502