Monocular Metric Depth Estimation

Monocular metric depth estimation aims to reconstruct accurate 3D depth maps from a single image, a challenging problem due to the inherent scale ambiguity in perspective projection. Recent research focuses on improving the accuracy and generalization of these estimations, exploring techniques like leveraging language descriptions, incorporating data from other sensors (e.g., radar), utilizing robot kinematics, and employing novel training strategies with synthetic data and multi-scale vision transformers. These advancements are crucial for applications in robotics, autonomous driving, and 3D scene understanding, enabling more robust and reliable perception systems.