Advancements in Machine Learning-Based Depth Estimation

Abstract

Obtaining depth information (Depth estimation) from one or several Red-Green-Blue (RGB) images remains a crucial and challenging task in computer vision, with significant implications for applications of robotics, medical imaging, autonomous driving, augmented reality (AR), and so on. Machine learning has revolutionized depth estimation, making it the leading approach in recent years. This paper thoroughly reviews machine learning-based depth estimation methods, encompassing the classification and comparison of frequently used datasets, emerging trends, and key evaluation metrics. The study evaluates six state-of-the-art (SOTA) methods. Some common evaluation metrics are utilized to assess their performance. The results indicate that incorporating diffusion models, internal discretization, and semi-supervised learning techniques markedly enhances the performance of depth estimation models. These innovations provide valuable perspectives and experiences for future research in the field. By highlighting the effectiveness of these advanced methods, this review provides a solid foundation for understanding current capabilities and guiding future developments in machine learning-based depth estimation.