Reinvention of the Imaging and Sensing Pipeline for AI

Existing AI camera systems typically integrate conventional cameras with computer vision models. However, these cameras are designed for human visual perception, and the models are trained on standard RGB images. This raises a fundamental question: is this configuration truly optimal for AI? Motivated by this question, we initiated the Imaging & Sensing Pipeline Optimization Project, based on the hypothesis that a more optimized pipeline must exist. Under this overarching concept, we have conducted a series of research studies, and in this presentation, I will showcase several of our previously proposed methods.