报告题目：高效的机器视觉系统：针对边缘图像传感器的电路-架构-算法联合设计

报 告 人：马天瑞

报告时间：2023年11月9日 9:00-10:00

报告地点：腾讯会议 621-749-206

摘要：The modern world craves for rich contextual information, much of which is driven by diverse vision applications. In these applications, humans are often the end-consumers of the images, therefore faithful capture and reconstruction of the original light scene become an important quality measure. Consequently, image sensors allocate the majority of their energy and latency budget to achieve higher image quality, resulting in significant energy/latency overheads in analog-to-digital conversion, image processing, and image transmission. Most importantly, these overheads scale with image resolution.

Recent accelerated advancements of deep learning-based computer vision have unleashed the second wave of machine vision. In this second wave, voluminous vision data are increasingly generated by intelligent edge image sensors and consumed, not by humans, but by downstream vision algorithms to perform sophisticated tasks such as classification, recognition, and machine perception. It presents a unique opportunity for innovative vision -- Given that images are destined for the downstream vision algorithms without the need for high-fidelity reconstruction, it is now possible to compress and preserve the "task-specific" information to reduce energy/latency overheads. Nonetheless, due to the constrained hardware resources on edge image sensors and the latency requirements of vision tasks, it is important to properly define and efficiently extract the task-specific information. Therefore, in this seminar, I show how I construct efficient machine vision systems with a co-design paradigm of circuit, architecture, and algorithm based on in-sensor computing.

报告人简介：

Tianrui Ma received the B.Eng. degree in Electrical Engineering (Hons.) from Beihang University, Beijing, China, and the M.S. degree in Electrical Engineering from Washington University in St. Louis, St. Louis, MO, USA, in 2019 and 2023, respectively. From July 2023 to October 2023, he was a research intern in the department of analog circuit design at OmniVision Technologies, Santa Clara, CA, USA. He is currently pursuing the Ph.D. degree in Electrical Engineering at Washington University in St. Louis.

His research interest focuses on CMOS image sensor-based machine learning hardware, electronic design automation, and computer architecture. He is the recipient of ISLPED 2022 Best Paper Award (BPA).