Congrats to Chengyu!
Title: Deep learning for camera autofocus
Abstract: To date, specialized autofocus sensors, such as phase detection or active ranging, are faster image-based autofocus algorithms. However, the need for specialized sensors and the assumption of the existence of a global “best focus” make focus sensors nonideal. This paper proposes a new pipeline for image- based autofocus and shows that neural image analysis finds focus 5-10x faster than traditional contrast enhancement. We achieve this by learning the direct mapping between an image and its focus position. In further contrast with conventional methods, AI methods can generate scene-based focus trajectories that optimize synthesized image quality for dynamic and three dimensional scenes. We propose a focus control strategy that captures a sequence of frames so that an all-in-focus image or video can be generated. In this strategy, a focus control agent analyzes the instantaneous estimation of the environment and determines the region of interest for focus analysis. We propose a rule-based agent and a learned agent for different scenarios and show their advantages over other focus stacking methods.