Skip to main navigation Skip to search Skip to main content

AFSN: Adaptive Frame Selection Network for Efficient Human Synthesis on Smartphones

  • Xuyang Feng
  • , Bin Guo
  • , Chen Qiu
  • , Haoyang Ren
  • , Zhiwen Yu
  • Northwestern Polytechnical University Xian

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Viewing real-time human synthesis results on smart-phones provides users a convenient interactive experience. How-ever, most human synthesis solutions relay on high-performance servers and lack interaction with users. In order to reduce traffic between smartphones and cloud, this paper introduces a novel adaptive frame selection network (AFSN) on mobile devices. Besides, we propose a reinforcement-learning based agent that selects the image frames according to the pose change, that reduces communications and the amount of calculations. Based upon AFSN, we build a multi-task human synthesis system, that can deal with background changing, pose transfer, virtual try-on, and other generate tasks in a unified framework. Experimental results on one real-world dataset demonstrate that our method can reduce the number of frames to be processed by more than 50%.

Original languageEnglish
Title of host publicationProceedings - 2022 8th International Conference on Big Data Computing and Communications, BigCom 2022
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages224-232
Number of pages9
ISBN (Electronic)9781665473842
DOIs
StatePublished - 2022
Event8th International Conference on Big Data Computing and Communications, BigCom 2022 - Xiamen, China
Duration: 6 Aug 20227 Aug 2022

Publication series

NameProceedings - 2022 8th International Conference on Big Data Computing and Communications, BigCom 2022

Conference

Conference8th International Conference on Big Data Computing and Communications, BigCom 2022
Country/TerritoryChina
CityXiamen
Period6/08/227/08/22

Keywords

  • cloud-device collaborative
  • human image synthesis
  • reinforcement learning

Fingerprint

Dive into the research topics of 'AFSN: Adaptive Frame Selection Network for Efficient Human Synthesis on Smartphones'. Together they form a unique fingerprint.

Cite this