Image Manipulation by Diffusion Models [in Russian]

Докладчик: Айбек Аланов (AIRI, НИУ ВШЭ)

Large-scale text-to-image diffusion models have shown their impressive ability to synthesize diverse and high-quality images. However, it is still challenging to directly apply these models for editing real images or generating special concepts provided by the user. Recently there were proposed many works to deal with these problems. In this seminar, we will examine most of these approaches, analyze its properties and indicate their advantages and weak sides. At first, we will start with a brief overview of contemporary text-to-image diffusion models. Secondly, we will give the formulation for two types of image manipulation problems: personalized generation of the user-provided concept and editing of the given real image. Then we will examine current methods that tackle these problems.

Статьи:
https://arxiv.org/abs/2208.01618
https://arxiv.org/abs/2208.12242
https://arxiv.org/abs/2208.01626
https://arxiv.org/abs/2210.09276
https://arxiv.org/abs/2211.09794