Revolutionizing Image Transformation: Unleashing the Power of Machine Learning for 2D to 3D Conversion

Abstract

This project offers a comprehensive evaluation of recent advancements in Neural Radiance Field (NeRF), a significant development in Computer Vision. NeRF models, known for their neural network-based scene representation and novel view synthesis, find applications in robotics, urban mapping, autonomous navigation, and virtual reality/augmented reality. The survey categorizes findings into architecture and application- based taxonomies, introducing NeRF theory and training via differentiable volume rendering. A benchmark compares key NeRF models' performance and speed, aiming to objectively assess strengths, weaknesses, and applicability across real-life scenarios. Results cover comparisons with photogrammetry in noise level, geometric accuracy, and image baselines across diverse scenes. NeRFs outperform in scenarios with texture-less, reflective, and refractive surfaces, while photogrammetry excels with cooperative textures.