Frequency regularized tensorial radiance field (free-TensoRF)

• Main Author: Ingale, Omkar
• Other Authors: Lu Shijian
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science
• Online Access: https://hdl.handle.net/10356/176001
• Institution: Nanyang Technological University

This project aims to enhance the rendering quality of Neural Radiance Field (NeRF) acceleration techniques, specifically focusing on Tensorial Radiance Field (TensoRF) which was introduced by Xu, Zexiang et al in 2022. NeRF, initially proposed by Mildenhall et al in 2020, revolutionized scene rendering through the use of a Multi-Layer Perceptron (MLP). While TensoRF demonstrated significant improvements over base NeRF by modelling scenes as a 4D tensor, there remains untapped potential for refining its rendering quality. In light of advancements in since TensoRF’s publication, this project seeks to explore innovative techniques, such as frequency regularization, to elevate the rendering quality of TensoRF. The primary objective is to improve TensoRF’s rendering quality without compromising its appealing attributes, such as rendering speed and memory efficiency. Through this exploration, the project aims to contribute to the ongoing evolution of NeRF acceleration techniques, pushing the boundaries of rendering quality and efficiency.