CNS Core: Small: Collaborative: Content-Based Viewport Prediction Framework for Live Virtual Reality Streaming

CNS 核心：小型：协作：用于直播虚拟现实流的基于内容的视口预测框架

• 批准号: 1909172
• 负责人: Yanzhi Wang
• 金额: $ 17.12万
• 依托单位: Northeastern University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-08-15 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1909172&HistoricalAwards=false
• 关键词: CNS; Core; Small; Collaborative; Content

Virtual reality (VR) video streaming has been gaining popularity recently with the rapid adoption of mobile head mounted display (HMD) devices in the consumer video market. As the cost for the immersive experience drops, VR video streaming introduces new bandwidth and performance challenges, especially in live streaming, due to the delivery of 360-degree views. This project develops a new content-based viewport prediction framework to improve the bandwidth and performance in live VR streaming, which predicts the user's viewport through a fusion of tracking the moving objects in the video, extracting the video semantics, and modeling the user's viewport of interest.This project consists of three research thrusts. First, it develops a content-based viewport prediction framework for live VR streaming by tracking the motions and semantics of the objects. Second, it employs hardware and software techniques to facilitate real-time execution and scale the viewport prediction mechanism to a large number of users. Third, it develops evaluation frameworks to verify the functionality, performance, and scalability of the approach. The project uniquely considers the correlation between video content and user behavior, which leverages the deterministic nature of the former to conquer the randomness of the latter.With the rapidly increasing popularity of VR systems in domain-specific immersive environments, the project will benefit several VR-related fields of studies with significant bandwidth savings and performance improvements, such as VR-based live broadcast, healthcare, and scientific visualization. Moreover, the interdisciplinary nature of the project will enhance the education and recruitment of underrepresented minorities in several science, technology, engineering, and mathematics (STEM) fields.The project repository will be stored on a publicly accessible server (https://github.com/hwsel). All the project data will be maintained for at least five years following the end of the grant period.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.

随着消费者视频市场中移动的头戴式显示器（HMD）设备的快速采用，虚拟现实（VR）视频流传输最近已经变得流行。随着沉浸式体验成本的下降，VR视频流引入了新的带宽和性能挑战，特别是在实时流媒体中，由于提供360度视图。本项目开发了一种新的基于内容的视口预测框架，通过跟踪视频中的移动对象、提取视频语义和建模用户感兴趣的视口来预测用户的视口，以提高实时VR流媒体的带宽和性能。本项目包括三个研究方向。首先，它通过跟踪对象的运动和语义，为实时VR流开发了一个基于内容的视口预测框架。其次，它采用硬件和软件技术来促进实时执行，并将视口预测机制扩展到大量用户。第三，它开发了评估框架，以验证该方法的功能，性能和可扩展性。该项目独特地考虑了视频内容和用户行为之间的相关性，利用前者的确定性来克服后者的随机性。随着VR系统在特定领域沉浸式环境中的迅速普及，该项目将通过显著的带宽节省和性能提升而受益于多个VR相关领域的研究，例如基于VR的直播，医疗保健，和科学可视化。此外，该项目的跨学科性质将加强在科学、技术、工程和数学（STEM）领域中代表性不足的少数群体的教育和招聘。项目库将存储在一个可公开访问的服务器上（https：//github.com/hwsel）。所有项目数据将在资助期结束后至少保留五年。该奖项反映了NSF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

Non-Structured DNN Weight Pruning--Is It Beneficial in Any Platform?

• DOI: 10.1109/tnnls.2021.3063265
• 发表时间: 2021-03-18
• 期刊: IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
• 影响因子: 10.4
• 作者: Ma, Xiaolong;Lin, Sheng;Wang, Yanzhi
• 通讯作者: Wang, Yanzhi

All-in-One: A Highly Representative DNN Pruning Framework for Edge Devices with Dynamic Power Management

• DOI: 10.1145/3508352.3549379
• 发表时间: 2022-10
• 期刊: 2022 IEEE/ACM International Conference On Computer Aided Design (ICCAD)
• 作者: Yifan Gong;Zheng Zhan;Pu Zhao;Yushu Wu;Chaoan Wu;Caiwen Ding;Weiwen Jiang;Minghai Qin;Yanzhi Wang

Teachers Do More Than Teach: Compressing Image-to-Image Models

• DOI: 10.1109/cvpr46437.2021.01339
• 发表时间: 2021-03
• 期刊: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
• 作者: Qing Jin;Jian Ren;Oliver J. Woodford;Jiazhuo Wang;Geng Yuan;Yanzhi Wang;S. Tulyakov

Layer Freezing & Data Sieving: Missing Pieces of a Generic Framework for Sparse Training

• DOI: 10.48550/arxiv.2209.11204
• 发表时间: 2022-09
• 期刊: ArXiv
• 作者: Geng Yuan;Yanyu Li;Sheng Li;Zhenglun Kong;S. Tulyakov;Xulong Tang;Yanzhi Wang;Jian Ren

A Unified DNN Weight Pruning Framework Using Reweighted Optimization Methods

使用重新加权优化方法的统一 DNN 权重修剪框架

• 发表时间: 2021
• 期刊: 58th ACM/IEEE Design Automation Conference (DAC
• 作者: Zhang, T.;Ma, X.;Zhan, Z.;Zhou, S.;Ding, C.;Fardad, M.;Wang, Y.
• 通讯作者: Wang, Y.