Citation

See, John Su Yang and Li, Yuxi and Lin, Weiyao and Xu, Ning and Xu, Shugong and Yan, Ke and Yang, Cong (2020) CFAD: Coarse-to-Fine Action Detector for Spatiotemporal Action Localization. In: CFAD: Coarse-to-Fine Action Detector for Spatiotemporal Action Localization. Springer Science and Business Media Deutschland GmbH, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), pp. 510-527. ISBN 9783030585167

Text 62.pdf - Published Version Restricted to Repository staff only Download (3MB)

Abstract

Most current pipelines for spatio-temporal action localization connect frame-wise or clip-wise detection results to generate action proposals, where only local information is exploited and the efficiency is hindered by dense per-frame localization. In this paper, we propose Coarseto-Fine Action Detector (CFAD), an original end-to-end trainable framework for efficient spatio-temporal action localization. The CFAD introduces a new paradigm that first estimates coarse spatio-temporal action tubes from video streams, and then refines the tubes’ location based on key timestamps. This concept is implemented by two key components, the Coarse and Refine Modules in our framework. The parameterized modeling of long temporal information in the Coarse Module helps obtain accurate initial tube estimation, while the Refine Module selectively adjusts the tube location under the guidance of key timestamps. Against other methods, the proposed CFAD achieves competitive results on action detection benchmarks of UCF101-24, UCFSports and JHMDB-21 with inference speed that is 3.3× faster than the nearest competitor.

Item Type:	Book Section
Uncontrolled Keywords:	Temporal automata
Subjects:	Q Science > QA Mathematics > QA150-272.5 Algebra
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	19 Aug 2021 15:09
Last Modified:	19 Aug 2021 15:09
URII:	http://shdl.mmu.edu.my/id/eprint/8254

Downloads

Edit (login required)