Speaker：Yukun Lai (Cardiff University)

Time：2023-11-21 16:30

Location：Conference Room 686 at the 6th floor of Shuli Building at Haiyun Campus

Abstract:

3D representations of objects including their shapes and appearance are fundamental for many applications such as virtual and augmented reality. It allows the objects to be freely viewed from different perspectives. However,  obtaining high quality 3D representations can often be time-consuming and require substantial expertise, e.g., to use dedicated tools such as 3D scanners. In this talk, I will present our recent work that aims to reduce the effort needed for high-quality 3D reconstruction by using machine learning to exploit prior knowledge. We further develop learning based methods to facilitate editing of 3D representations, in particular neural radiance fields, to intuitively modify the geometry and/or appearance. I will discuss how specific techniques can help address challenges for different tasks, along with some directions for future research.