Evgenii Kashin

University of York

Evgenii, a Computer Science enthusiast, began crafting games in school using the Warcraft3 editor. He spent five years as a Machine Learning Engineer, excelling in computer vision and graphics. His work at Snap included creating engaging lenses and researching 3D object capturing. An ECCV2020 article on face manipulation, with over 100 citations, is a testament to his prowess. Away from work, he enjoys bouldering, hiking, racing, and gaming.     

My research is dedicated to establishing a cost-effective approach for creating and generating 3D scenes for game development, a critical aspect of modern VR/AR applications. Harnessing the potential of generative visual content, I aim to develop algorithms capable of realistically completing 3D scenes from a few images. This could revolutionize the entertainment and creative industries, particularly game development. Picture having only a couple of images from your favourite film and envisioning the entire scene. Such technology can enhance the efficiency of 3D artists, democratize game development, and serve as entertainment in itself. Currently, I am developing an algorithm to achieve this goal. The proposed solution employs a general pretrained text-to-image model for supervision, with a NeRF 3D representation of the scene. The central concept involves iterative outpainting, where each iteration updates the NeRF weights.