Visit ShanghaiTech University | 中文 | How to find us
HOME > News and Events > Events
A 58.6mW Real-Time Programmable Object Detector with Multi-Scale Multi-Object?Support Using Deformable Parts Model on 1920x1080 Video at 30fps
Date: 2016/6/21             Browse: 306

A 58.6mW Real-Time Programmable Object Detector with Multi-Scale Multi-Object?Support Using Deformable Parts Model on 1920x1080 Video at 30fps

Speaker: Zhengdong Zhang

Time: Jun 21, 10:00am - 11:00am.

Location: Room 316, H2 Building

Abstract:

Object detection from visual data like images and videos has seeing increasing applications in surveillance, advanced driver assistant system (ADAS), autonomous cars and augmented realities. However, existing object detection algorithms are slow on high resolution data; they also consume a lot of power, which prohibits them from being deployed to power constrained systems like mobile phones and embedded devices.

In this talk, we present a programmable, energy-efficient and real-time object detection chip using deformable parts models (DPM), with 2x higher accuracy than traditional rigid body models. With 8 deformable parts detection, three methods are used to address the high computational complexity: classification pruning for 33x fewer parts classification, vector quantization for 15x memory size reduction, and basis projection for 2x reduction of the cost of each classification.

The chip is implemented using 65nm technology, and can process 1920x1080 images in real-time at 30fps without any off-chip storage while consuming only 58.6mW (0.94nJ/pixel, 1168 GOPS/W). The chip is programmable and can detect 2 different classes of objects simultaneously. It is energy scalable by changing the pruning factor or disabling the parts classification.

Such an energy efficient chip enables long-time non-stop object detection on embedded devices with limited battery capacity. It makes the camera significantly smarter and allows its intelligence to be used at much larger scale.

Bio:

Zhengdong Zhang received the B.S. in Computer Science in 2011 from Tsinghua University, Beijing, China. He received the M.S. degree in Computer Science from Massachusetts Institute of Technology, Cambridge in 2014. Between 2011 and 2012 he worked in Microsoft Research Asia, Beijing, China, as an Assistant Researcher. He is pursuing the Ph.D. degree under the supervision of Prof. Vivienne Sze. His research interest spans the area of sparsity, low-rank matrix recovery, symmetry/regularity of textures, 3D computer vision, computational photography, video coding, super-resolution. His current research focuses on the design of energy-efficient vision systems.

SIST-Seminar 16047