End-to-end optimization of meta-lens doublet for high-quality wide-angle imaging

Abstract

In this paper, we propose an advanced method to jointly optimize doublet metalens and deep learning-based postprocessing networks for wide-angle and full-color imaging with high fidelity. The optical image formation module in the spatially-variant system and a reconstruction network module are implemented in a differentiable manner. By premitigating coma aberration with doublet metalens, the proposed model outperforms both cases of singlet structure and analogous electronic implementations in terms of reconstruction accuracy.