< Back to previous page

Project

From images to photorealistic renderings

Given (a) real image(s) of an object or scene, the goal of this project is to generate a virtual model that allows rendering the same object or scene from different viewpoints or under different illumination conditions. This requires estimating, implicitly or explicitly, the different image intrinsics: geometry (3D), reflectance (texture), and shading (illumination). It also involves 'inpainting' or 'hallucinating' the parts of the object that are (self)occluded in the input images. This will be approached in a data-driven manner, leveraging deep learning and generative models, for specific object categories such as cars.

Date:15 Oct 2021 →  Today
Keywords:computer vision, real2sim, rendering
Disciplines:Computer vision
Project type:PhD project