Toward practical depth estimation based on deep learning

Depth estimation from images is a classic computer vision problem that has occupied researchers for decades. Obtaining dense and accurate depth estimation is pivotal to effectively address higher-level tasks in computer vision such as autonomous driving, 3D reconstruction, and robotics. Therefore, with a focus on diverse depth estimation setups, this thesis addresses significant challenges that impact methodologies for inferring depth from images or obtaining completed depth from hardware sensors. The goal is to develop precise and efficient frameworks tailored for accurate Simultaneous Localization And Mapping (SLAM) in challenging environments.