This code implements a differentiable proxy model for simulating meta-optics and a neural feature propagation deconvolution method. These components are optimized end-to-end using machine learning ...