Neural Scene Representations for Learning-based View Synthesis and Its Applications
Phong Nguyen-Ha received the B.Sc. degree in mechanical engineering from the Ha Noi University of Science and Technology (HUST), Vietnam and the M.Sc. degree in computer science engineering at Dongguk University, Seoul, South Korea. Since then, he has been working as a doctoral candidate at University of Oulu, Finland, fully funded by the Vision-based 3D perception for mixed reality applications. His research interests include 3D computer vision, computer graphics and deep learning. During his PhD, he also worked at Meta and Nvidia as research scientist interns.
Creating lifelike images and videos has been a major research focus in computer graphics for many years. In the past, generating synthetic images required using specific 3D representations of the geometry and material properties, which were processed through algorithms such as rasterization or ray tracing. However, working with real-world data is more challenging as traditional representations like voxels or point clouds can be computationally intensive. Neural scene representations, on the other hand, are more streamlined and efficient, making them faster and more effective. Additionally, neural scene representations can be trained end-to-end from data, allowing them to be tailored to particular tasks and domains, including novel view synthesis, relighting, scene composition, text-to-3D, text-to-4D, and more.