Exploiting nonlinear incoherent image formation through linear volume metaoptics for inference [Invited]

We showed that a 2D depth map representing an incoherent 3D opaque scene is directly encoded in the response function of an imaging optics. As a result, the optics creates an image that depends nonlinearly on the depth map. Furthermore, strong spatio-spectral dispersions in volume metaoptics can be engineered to create an arbitrarily complex image in response to a depth map. We hypothesize that this complexity will allow the linear volume metaoptics to directly sense and process 3D opaque scenes for deep vision tasks.