Neural array meta-imaging

Abstract Compact, high-quality imaging systems are highly desired for scientific, industrial, and consumer applications. Metalenses combined with computational imaging offer a promising solution for developing such systems, yet their performance is fundamentally limited by the commonly used point-to-point imaging model, which forces trade-offs between aperture size, F-number, field of view (FOV), waveband width, and image quality. Here, we experimentally demonstrate that a neural array imaging model can overcome these long-standing trade-offs, achieving a 25-Hz full-color imaging camera with a 2.76-mm aperture, 1.45 F-number, 50 $$^{\circ }$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mmultiscripts> <mml:mrow/> <mml:mrow/> <mml:mo>∘</mml:mo> </mml:mmultiscripts> </mml:math> FOV, and a spectral range of 400–700 nm. The camera achieves image quality comparable to commercial compound lenses (e.g., Edmund 33-300) in both indoor and outdoor environments, while reducing the total track length by a factor of 13. We further demonstrate its suitability for object detection and depth estimation in real-world scenarios. This neural array imaging model is also applied to polarization imaging, showcasing its scalability and versatility for broadband applications.