Scene Geometry

Scene geometry research focuses on accurately representing and reconstructing the three-dimensional structure of scenes from various visual inputs, such as images, videos, and LiDAR scans. Current research emphasizes developing novel neural representations, including radiance fields (NeRFs) and Gaussian splatting, often incorporating depth information and leveraging techniques like point clouds and signed distance functions to improve accuracy and efficiency. These advancements have significant implications for applications like autonomous driving, robotics, virtual and augmented reality, and computer graphics, enabling more realistic and robust scene understanding and manipulation.