NeRF Explored: A Comprehensive Analysis of Neural Radiance Field in 3D Vision
Muhammad Haider, National University of Sciences and Technology (NUST), Islamabad, Pakistan.
Khurram Shahzad, National University of Sciences and Technology (NUST), Islamabad, Pakistan.
Ammad Ullah Rameez, National University of Sciences and Technology (NUST), Islamabad, Pakistan.
Sajid Umair, University of Missouri Kansas City (UMKC), USA.
Safdar Abbas Khan, National University of Sciences and Technology (NUST), Islamabad, Pakistan.
Corresponding Author:
Muhammad Haider (mhaider.mscs20seecs@seecs.edu.pk)
Abstract:
Neural Radiance Fields (NeRF) refers to a method in computer vision that uses the power of Neural Networks to synthesize three-dimensional scenes from two-dimensional images. They enable the creation of detailed and realistic images after modeling three-dimensional scene as a function that predicts color and opacity at any given point in space. This allows the creation of new views from any desired point. NeRFs are therefore very useful in virtual/ augmented reality and computer-generated imagery. Using Neural Networks, NeRFs achieve unparalleled fidelity showing its importance in technology innovation. This paper provides an overview of NeRF highlighting its evolution and uses in diverse applications across various domains such as robotics navigation, urban reconstruction, human face generation, and many more. Our work provides readers with a systematic examination of NeRF's advancements by delving into different approaches and architecture used in the literature and will provide key insights to both novice and seasoned researchers in this dynamic field of computer vision.
Keywords:
Neural Radiance Field; NeRF, 3D; Deep Learning; View Synthesis; Neural Networks; Multi-Layer Perceptron.