Inverse Design of Cascaded Multidimensional Coding Metasurface for Snapshot Full-Stokes Imaging and Depth Estimation

Maximizing channel capacity while contending with a limited spatiotemporal bandwidth has consistently presented a critical trade-off in the simultaneous acquisition of multidimensional light field information. The codesign of optical coding elements and reconstruction algorithms offers a promising avenue to address this challenge. However, existing optical coding elements typically only encode one single dimension, as compact multidimensional encoding imposes higher demands on both the coding hardware and the recovery algorithms. Herein, we propose a cascaded multidimensional coding metasurface for snapshot polarization imaging and monocular depth estimation. Leveraging a data-driven framework, an imaging, depth-coding metasurface and a nonimaging, polarization-coding metasurface are codesigned alongside a backend reconstruction algorithm. This joint optimization achieves a compact depth-polarization encoding–decoding system and significantly enhances the detector pixel utilization efficiency. Experimentally, our method enables the acquisition of precise full-Stokes images and depth maps. This approach demonstrates the design capabilities of end-to-end computational imaging frameworks for multiple optical devices and further highlights the potential of metasurfaces as optical coding elements for multidimensional light field information acquisition.