• 文献检索
  • 文档翻译
  • 深度研究
  • 学术资讯
  • Suppr Zotero 插件Zotero 插件
  • 邀请有礼
  • 套餐&价格
  • 历史记录
应用&插件
Suppr Zotero 插件Zotero 插件浏览器插件Mac 客户端Windows 客户端微信小程序
定价
高级版会员购买积分包购买API积分包
服务
文献检索文档翻译深度研究API 文档MCP 服务
关于我们
关于 Suppr公司介绍联系我们用户协议隐私条款
关注我们

Suppr 超能文献

核心技术专利：CN118964589B侵权必究
粤ICP备2023148730 号-1Suppr @ 2026

文献检索

告别复杂PubMed语法，用中文像聊天一样搜索，搜遍4000万医学文献。AI智能推荐，让科研检索更轻松。

立即免费搜索

文件翻译

保留排版，准确专业，支持PDF/Word/PPT等文件格式，支持 12+语言互译。

免费翻译文档

深度研究

AI帮你快速写综述，25分钟生成高质量综述，智能提取关键信息，辅助科研写作。

立即免费体验

在移动设备上使用异构处理器的以感知为中心的图像超分辨率技术

Perceptual-Centric Image Super-Resolution using Heterogeneous Processors on Mobile Devices.

作者信息

Huang Kai, Yin Xiangyu, Gu Tao, Gao Wei

机构信息

University of Pittsburgh, USA.

Macquarie University, Australia.

出版信息

Proc Annu Int Conf Mob Comput Netw. 2024 Nov;2024:1361-1376. doi: 10.1145/3636534.3690698. Epub 2024 Dec 4.

DOI:10.1145/3636534.3690698
PMID:40129861
原文链接:https://pmc.ncbi.nlm.nih.gov/articles/PMC11931654/
Abstract

Image super-resolution (SR) is widely used on mobile devices to enhance user experience. However, neural networks used for SR are computationally expensive, posing challenges for mobile devices with limited computing power. A viable solution is to use heterogeneous processors on mobile devices, especially the specialized hardware AI accelerators, for SR computations, but the reduced arithmetic precision on AI accelerators can lead to degraded perceptual quality in upscaled images. To address this limitation, in this paper we present , a novel image SR technique that enhances the perceptual quality of upscaled images when using heterogeneous processors for SR computations. FYE-SR strategically splits the SR model and dispatches different layers to heterogeneous processors, to meet the time constraint of SR computations while minimizing the impact of AI accelerators on image quality. Experiment results show that FYE-SR outperforms the best baselines, improving perceptual image quality by up to 2×, or reducing SR computing latency by up to 5.6× with on-par image quality.

摘要

图像超分辨率（SR）广泛应用于移动设备以提升用户体验。然而，用于SR的神经网络计算成本高昂，给计算能力有限的移动设备带来了挑战。一个可行的解决方案是在移动设备上使用异构处理器，特别是专用硬件AI加速器来进行SR计算，但AI加速器上降低的算术精度可能会导致放大图像的感知质量下降。为解决这一限制，在本文中我们提出了一种新颖的图像SR技术，当使用异构处理器进行SR计算时，该技术可提升放大图像的感知质量。FYE - SR策略性地拆分SR模型并将不同层分配到异构处理器，以满足SR计算的时间约束，同时将AI加速器对图像质量的影响降至最低。实验结果表明，FYE - SR优于最佳基线，在图像质量相当的情况下，将感知图像质量提高了2倍，或者将SR计算延迟降低了5.6倍。

https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/f9014a11d7cf/nihms-2065959-f0024.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/4d60640f83fb/nihms-2065959-f0001.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/b6047658c7fc/nihms-2065959-f0002.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/99c64819363c/nihms-2065959-f0003.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/aba888e2c268/nihms-2065959-f0004.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/4451b36d76a3/nihms-2065959-f0005.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/5b4453fd8f4e/nihms-2065959-f0006.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/73898ee5edf4/nihms-2065959-f0007.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/ffda632d5008/nihms-2065959-f0008.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/c0566aeb1589/nihms-2065959-f0009.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/d2978616879f/nihms-2065959-f0010.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/d19a69517149/nihms-2065959-f0011.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/e2b0698f8bd1/nihms-2065959-f0012.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/2cdab2be1ba9/nihms-2065959-f0013.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/124e685ef8f6/nihms-2065959-f0014.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/5713befbb329/nihms-2065959-f0015.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/4cd022ac5d2a/nihms-2065959-f0016.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/ba5036e18264/nihms-2065959-f0017.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/9f1ca4982d59/nihms-2065959-f0018.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/9d3d75d32b65/nihms-2065959-f0019.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/3d226edf976c/nihms-2065959-f0020.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/8e7364048fbd/nihms-2065959-f0021.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/b87984c74123/nihms-2065959-f0022.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/e2ed9bcae25d/nihms-2065959-f0023.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/f9014a11d7cf/nihms-2065959-f0024.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/4d60640f83fb/nihms-2065959-f0001.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/b6047658c7fc/nihms-2065959-f0002.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/99c64819363c/nihms-2065959-f0003.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/aba888e2c268/nihms-2065959-f0004.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/4451b36d76a3/nihms-2065959-f0005.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/5b4453fd8f4e/nihms-2065959-f0006.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/73898ee5edf4/nihms-2065959-f0007.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/ffda632d5008/nihms-2065959-f0008.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/c0566aeb1589/nihms-2065959-f0009.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/d2978616879f/nihms-2065959-f0010.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/d19a69517149/nihms-2065959-f0011.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/e2b0698f8bd1/nihms-2065959-f0012.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/2cdab2be1ba9/nihms-2065959-f0013.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/124e685ef8f6/nihms-2065959-f0014.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/5713befbb329/nihms-2065959-f0015.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/4cd022ac5d2a/nihms-2065959-f0016.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/ba5036e18264/nihms-2065959-f0017.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/9f1ca4982d59/nihms-2065959-f0018.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/9d3d75d32b65/nihms-2065959-f0019.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/3d226edf976c/nihms-2065959-f0020.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/8e7364048fbd/nihms-2065959-f0021.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/b87984c74123/nihms-2065959-f0022.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/e2ed9bcae25d/nihms-2065959-f0023.jpg
https://cdn.ncbi.nlm.nih.gov/pmc/blobs/8134/11931654/f9014a11d7cf/nihms-2065959-f0024.jpg

相似文献

1
Perceptual-Centric Image Super-Resolution using Heterogeneous Processors on Mobile Devices.在移动设备上使用异构处理器的以感知为中心的图像超分辨率技术
Proc Annu Int Conf Mob Comput Netw. 2024 Nov;2024:1361-1376. doi: 10.1145/3636534.3690698. Epub 2024 Dec 4.
2
UArch: A Super-Resolution Processor With Heterogeneous Triple-Core Architecture for Workloads of U-Net Networks.UArch：一种具有异构三核架构的超分辨率处理器，适用于 U-Net 网络的工作负载。
IEEE Trans Biomed Circuits Syst. 2023 Jun;17(3):633-647. doi: 10.1109/TBCAS.2023.3261060. Epub 2023 Jul 12.
3
Super-resolution reconstruction of ultrasound image using a modified diffusion model.基于改进扩散模型的超声图像超分辨率重建。
Phys Med Biol. 2024 Jun 17;69(12). doi: 10.1088/1361-6560/ad4086.
4
A novel hybrid generative adversarial network for CT and MRI super-resolution reconstruction.一种用于 CT 和 MRI 超分辨率重建的新型混合生成对抗网络。
Phys Med Biol. 2023 Jun 23;68(13). doi: 10.1088/1361-6560/acdc7e.
5
SGSR: style-subnets-assisted generative latent bank for large-factor super-resolution with registered medical image dataset.SGSR：基于风格子网的生成式潜在银行，用于具有注册医学图像数据集的大因子超分辨率。
Int J Comput Assist Radiol Surg. 2024 Mar;19(3):493-506. doi: 10.1007/s11548-023-03037-3. Epub 2023 Dec 21.
6
Super-Interpolation With Edge-Orientation-Based Mapping Kernels for Low Complex 2× Upscaling.基于边缘定向映射核的超插值用于低复杂度的 2×上采样。
IEEE Trans Image Process. 2016 Jan;25(1):469-83. doi: 10.1109/TIP.2015.2507402. Epub 2015 Dec 9.
7
Patient-specific MRI super-resolution via implicit neural representations and knowledge transfer.通过隐式神经表示和知识转移实现针对特定患者的磁共振成像超分辨率
Phys Med Biol. 2025 Apr 1;70(7). doi: 10.1088/1361-6560/adbed4.
8
Enhancing Ultrasound Image Quality Across Disease Domains: Application of Cycle-Consistent Generative Adversarial Network and Perceptual Loss.跨疾病领域提升超声图像质量：循环一致生成对抗网络与感知损失的应用
JMIR Biomed Eng. 2024 Dec 17;9:e58911. doi: 10.2196/58911.
9
Reparameterizable Multibranch Bottleneck Network for Lightweight Image Super-Resolution.可重参数化多分支瓶颈网络的轻量级图像超分辨率。
Sensors (Basel). 2023 Apr 13;23(8):3963. doi: 10.3390/s23083963.
10
Super resolution-based methodology for self-supervised segmentation of microscopy images.基于超分辨率的显微镜图像自监督分割方法。
Front Microbiol. 2024 Mar 12;15:1255850. doi: 10.3389/fmicb.2024.1255850. eCollection 2024.

本文引用的文献

1
Deep Learning Single-Frame and Multiframe Super-Resolution for Cardiac MRI.深度学习单帧和多帧心脏 MRI 超分辨率。
Radiology. 2020 Jun;295(3):552-561. doi: 10.1148/radiol.2020192173. Epub 2020 Apr 14.
2
Deep Learning for Image Super-Resolution: A Survey.用于图像超分辨率的深度学习：一项综述。
IEEE Trans Pattern Anal Mach Intell. 2021 Oct;43(10):3365-3387. doi: 10.1109/TPAMI.2020.2982166. Epub 2021 Sep 2.
3
Image Super-Resolution Using Deep Convolutional Networks.基于深度卷积网络的图像超分辨率重建。
IEEE Trans Pattern Anal Mach Intell. 2016 Feb;38(2):295-307. doi: 10.1109/TPAMI.2015.2439281.