High-throughput and accurate cross-scale 3D sensing with a meta-depth sensor

Depth sensing has been widely applied in industrial inspection, transportation and consumer electronics. However, existing methods still face an inherent trade-off between depth precision and data throughput. Here, we present a meta-imaging-empowered depth sensor (MDS), a fully passive and monocular modality, that transcends this limitation through aberration-robust disparity-depth calibration and vision-foundation-model-empowered fine-grained three-dimensional (3D) ranging based on a scanning light-field imaging system. Our system synergizes high-throughput disparity estimation and massive data priors from foundation models, enabling high-precision, fine-grained metric depth estimation across depth ranges from 0.1 to 500 m. Experimental validation demonstrates 2.5-mm absolute depth precision at a 1.5-m distance (0.18% relative error) and 3-megapixel resolution, outperforming conventional light-field depth sensors by 7 times in accuracy and 5 times in spatial resolution without sacrificing imaging speed with a virtual-scanning technique. We also achieve turbulence-robust remote depth sensing at the 500-m distance with a significantly denser spatial sampling rate compared with commercial LiDAR systems, and a depth estimation precision of 0.7 m at a 235-m distance. By unifying the throughput advantages of passive imaging with the high-precision depth estimation ability of physics-aware neural regression, the proposed MDS establishes a scalable paradigm for high-fidelity passive 3D sensing across scales, with applications ranging from industrial inspection to autonomous driving.