Collaborative Research: SHF: Small: Enabling Efficient 3D Perception: An Architecture-Algorithm Co-Design Approach

协作研究：SHF：小型：实现高效的 3D 感知：架构-算法协同设计方法

• 批准号: 2126642
• 负责人: Yuhao Zhu
• 金额: $ 28.5万
• 依托单位: University of Rochester
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2126642&HistoricalAwards=false
• 关键词: Collaborative; Research; SHF; Small; Enabling

This research project designs software-hardware collaborative mechanisms to boost the execution efficiency of 3D perception (point cloud) algorithms by an order of magnitude. Achieving this goal requires designing fundamentally new algorithmic primitives, offline and run-time systems, and hardware architectures that tame the irregular computation and memory patterns in 3D perception algorithms. This research unlocks next-generation software innovation in emerging domains driven by 3D perception, such as autonomous driving, Augmented/Virtual Reality, and precision agriculture. The research agenda is complemented by an educational/outreach agenda. The PIs are (1) offering summer “introduction to computing” courses to high school students from the Rochester Central School District, (2) engaging students in RIT’s National Technical Institute for the Deaf program through experiencing 3D sensing and research activities, (3) introducing new courses/modules in both UR and RIT on 3D perception, both on algorithms and hardware systems, and (4) offering undergraduate students inclusive opportunities for hands-on experience in emerging application domains and hardware acceleration.This research project addresses the fundamental mismatch between the irregularities in point-cloud algorithms and today’s hardware architectures, which are primarily optimized for 2D image- and video-processing algorithms that are regular stencil pipelines operating on structured data. The key intellectual merit is the pursuit of new algorithms and system architectures that reduce/eliminate irregular computation and memory accesses in 3D perception. The technical contribution is three-fold: 1) efficient, yet generally applicable, hardware building blocks required to accelerate point cloud algorithms, 2) run-time systems that dynamically adapt to operating constraints (e.g., hardware resources, performance, energy) in an application-aware and data-aware manner, and 3) a new class of efficient-by-construction 3D perception algorithms that leverage the small data volume in single-beam point clouds. The algorithm-hardware co-designed system not only accelerates current 3D perception algorithms, but also provides a computing substrate so that 3D perception can be pervasively used as a building block in future application domains.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

本研究项目设计软硬件协同机制，将三维感知（点云）算法的执行效率提高一个数量级。实现这一目标需要从根本上设计新的算法原语、离线和运行时系统，以及能够驯服3D感知算法中不规则计算和内存模式的硬件架构。这项研究开启了由3D感知驱动的新兴领域的下一代软件创新，如自动驾驶、增强/虚拟现实和精准农业。研究议程由教育/外联议程补充。这些pi(1)为罗切斯特中央学区的高中生提供暑期“计算机入门”课程，(2)让学生参与RIT的国家聋人技术研究所项目，体验3D传感和研究活动，(3)在RIT和RIT引入新的3D感知课程/模块，包括算法和硬件系统。(4)为本科生提供在新兴应用领域和硬件加速方面的实践经验的机会。该研究项目解决了点云算法中的不规则性与当今硬件架构之间的根本不匹配，硬件架构主要针对二维图像和视频处理算法进行了优化，这些算法是在结构化数据上运行的常规模板管道。关键的智力优势是追求新的算法和系统架构，以减少/消除3D感知中的不规则计算和内存访问。其技术贡献有三方面：1)加速点云算法所需的高效且普遍适用的硬件构建块；2)以应用感知和数据感知方式动态适应操作约束（例如硬件资源、性能、能源）的运行时系统；3)利用单光束点云中的小数据量的新型高效3D感知算法。算法-硬件协同设计的系统不仅加速了当前的3D感知算法，而且提供了一个计算基础，使3D感知可以在未来的应用领域中广泛使用作为构建块。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Digital Reconstruction of Elmina Castle for Mobile Virtual Reality via Point-based Detail Transfer

• DOI: 10.2352/ei.2022.34.1.vda-409
• 发表时间: 2020-12
• 期刊: ArXiv
• 作者: Sifan Ye;Ting-Wei Wu;Michael Jarvis;Yuhao Zhu

RTNN: accelerating neighbor search using hardware ray tracing

• DOI: 10.1145/3503221.3508409
• 发表时间: 2022-01
• 期刊: Proceedings of the 27th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming
• 作者: Yuhao Zhu

ImaGen: A General Framework for Generating Memory- and Power-Efficient Image Processing Accelerators

• DOI: 10.1145/3579371.3589076
• 发表时间: 2023-04
• 期刊: Proceedings of the 50th Annual International Symposium on Computer Architecture
• 作者: Nisarg Ujjainkar;Jingwen Leng;Yuhao Zhu

Crescent: taming memory irregularities for accelerating deep point cloud analytics

• DOI: 10.1145/3470496.3527395
• 发表时间: 2022-04
• 期刊: Proceedings of the 49th Annual International Symposium on Computer Architecture
• 作者: Yu Feng;Gunnar Hammonds;Yiming Gan;Yuhao Zhu