A Graphics and Vision Unified Processor with 0.89µW/fps Pose Estimation Engine for Augmented Reality

📋 论文概要

该论文提出了一种面向增强现实的图形与视觉统一处理器，集成了姿态估计引擎，能够在低功耗下同时处理图形渲染和视觉特征提取。通过6路VLIW数据通路、可重构处理单元以及像素排列器，解决了传统分离架构效率低的问题。

💡 主要创新点

• 提出6路VLIW数据通路设计，支持图形与视觉指令并行执行
• 设计可重构处理单元，可在图形模式和视觉模式间切换，提高硬件利用率
• 引入像素排列器，其功能与图形光栅化相反，用于高效处理视觉像素数据

🏷 关键词

📄 原文摘要

Graphics processing generates pixels from descriptors, while vision generates descriptors from pixels [1]. Since augmented reality (AR) requires both graphics and vision abilities (Fig. 18.6.1), we report a unified processor for graphics, vision, and pose estimation for marker tracking. To process both graphics and vision simultaneously, we use 4 key features: (1) 6-way VLIW datapath design of processing elements, (2) reconfigurable processing elements for graphics and vision modes, (3) a pixel arranger for vision processing that has the inverse characteristic of a graphics rasterizer, and (4) a dedicated pose-estimation engine to generate graphics control data from vision processing. Using these methods, we achieve 371.9GOPS/W for full operation in a VGA image and

👥 作者与机构

Jae-Sung Yoon, Jeong-Hyun Kim, Hyo-Eun Kim, Won-Young Lee,

Seok-Hoon Kim, Kyusik Chung, Jun-Seok Park, Lee-Sup Kim KAIST, Daejeon, Korea In many ways, 3D graphics and vision processing are inverse operations