Convolutional neural networks are crucial for a variety of computer vision tasks, including object recognition, tracking, classification, and segmentation. However, the computational demands of these networks often result in substantial energy consumption and latency issues. Nanophotonic optical encoders, which leverage point spread functions acting as kernels, offer a solution by performing convolution in the optical domain. This innovative approach has led to over two orders of magnitude reduction in energy usage and four orders of magnitude improvement in latency, with ongoing efforts to expand its applications.