Cohere Labs - Yingsi Qin, PhD candidate, CMU

Conventional Autofocus in today's cameras can only focus to a single depth at a time, with a limited, planar depth of field. We propose Spatially-Varying Autofocus, a technique that can focus different parts of the sensor onto different depths in the scene, enabling an arbitrary focal surface while maintaining a large aperture and the highest possible spatial resolution. We realize this using a computational lens formed by a Lohmann lens and a phase-only spatial light modulator, together with spatially varying autofocus algorithms that iteratively estimate the depth map from contrast and disparity cues. Our technique allows the camera to progressively shape its depth-of-field to the scene geometry. In contrast to focus stacking, which relies on multi-shot capture, and lightfield cameras, which rely on post-capture processing, our real-time prototype optically streams all-in-focus videos at 21 FPS for dynamic scenes.

Yingsi is a PhD candidate in Electrical and Computer Engineering at Carnegie Mellon University, advised by Aswin Sakaranarayanan and Matthew O'Toole. She is a part of the Image Science Lab and Carnegie Mellon Computational Imaging.

Her research focuses on designing and building intelligent spatial optical systems for next-generation computational imaging, immersive displays, and embodied vision. Her research area involves a fusion of computer vision, optics, signal processing, and machine learning.

She obtained her B.S. in Computer Science from Columbia University, and my B.A. in Physics from Colgate University. I was a research intern at Meta Reality Labs Display Systems Research (2024, 2025) and Snap Research Computational Imaging (2020). She was also a software engineering intern at Google Search (2019).

Add event to calendar

Apple Google Office 365 Outlook Outlook.com Yahoo