Neural Network based VLSI architecture for HEVC Motion Estimation

用于 HEVC 运动估计的基于神经网络的 VLSI 架构

• 批准号: 17J10356
• 负责人: 王 世豪
• 金额: $ 1.09万
• 依托单位: Waseda University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for JSPS Fellows
• 财政年份: 2017
• 资助国家: 日本
• 起止时间: 2017-04-26 至 2018-03-31
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-17J10356/
• 关键词: CNN; Deep Learning

This research focuses on using newest deep learning algorithms and its implementation to improve motion estimation in video coding as it tightly affects the video coding quality and the energy consumption.The first year focuses on the hardware design part, which is to design an energy-efficient hardware architecture of deep learning algorithms. Specifically, the deep convolution neural networks are chosen for their superior performance on many image/video related tasks.I put efforts on two things, the scheduling algorithm for computing order and new hardware architectures. Both are regarded as the important issues in this fields as they affects on energy costs, which is critical for mobile devices. Both works are accepted in international conferences.

本研究的重点是使用最新的深度学习算法及其实现来改善视频编码中的运动估计，因为它密切影响视频编码质量和能耗。第一年的重点是硬件设计部分，即设计一个节能的深度学习算法硬件架构。具体来说，选择深度卷积神经网络是因为它们在许多图像/视频相关任务上的上级性能。我在两件事上付出了努力，计算顺序的调度算法和新的硬件架构。两者都被认为是该领域的重要问题，因为它们影响能源成本，这对移动的设备至关重要。这两部作品都被国际会议接受。

项目成果

IEEE International Symposium on Circuit and Systems (ISCAS), 2017

IEEE 国际电路与系统研讨会 (ISCAS)，2017

• 发表时间: 2017

Chain-NN: Energy-Efficient 1D Chain Architecture and Dataflow for Accelerating Deep Convolutional Neural Networks

Chain-NN：用于加速深度卷积神经网络的节能一维链架构和数据流

• 发表时间: 2018
• 期刊: IEEE Transactions on Circuits and Systems for Video Technology
• 影响因子: 8.4
• 作者: Shihao Wang;Dajiang Zhou;Xushen Han;Takeshi Yoshimura and Satoshi Goto
• 通讯作者: Takeshi Yoshimura and Satoshi Goto

Energy-efficient Scheduling Method with Cross-loop Model for customized CNN Accelerators,

定制 CNN 加速器的跨环模型节能调度方法，

• 发表时间: 2017
• 作者: Kaiyi Yang;Shihao Wang;Jianbin Zhou and Takeshi Yoshimura
• 通讯作者: Jianbin Zhou and Takeshi Yoshimura

Chain-NN: An energy-efficient 1D chain architecture for accelerating deep convolutional neural networks

• DOI: 10.23919/date.2017.7927142
• 发表时间: 2017-03
• 期刊: Design, Automation & Test in Europe Conference & Exhibition (DATE), 2017
• 作者: Shihao Wang;Dajiang Zhou;Xushen Han;T. Yoshimura