Dynamical hyperparameter optimization via deep reinforcement learning in tracking

dc.contributor.authorDong, Xingping
dc.contributor.authorShen, Jianbing
dc.contributor.authorWang, Wenguan
dc.contributor.authorShao, Ling
dc.contributor.authorLing, Haibin
dc.contributor.authorPorikli, Fatih
dc.date.accessioned2024-05-13T00:03:38Z
dc.date.issued2021
dc.date.updated2023-01-15T07:16:58Z
dc.description.abstractHyperparameters are numerical pre-sets whose values are assigned prior to the commencement of a learning process. Selecting appropriate hyperparameters is often critical for achieving satisfactory performance in many vision problems, such as deep learning-based visual object tracking. However, it is often difficult to determine their optimal values, especially if they are specific to each video input. Most hyperparameter optimization algorithms tend to search a generic range and are imposed blindly on all sequences. In this paper, we propose a novel dynamical hyperparameter optimization method that adaptively optimizes hyperparameters for a given sequence using an action-prediction network leveraged on continuous deep Q-learning. Since the observation space for object tracking is significantly more complex than those in traditional control problems, existing continuous deep Q-learning algorithms cannot be directly applied. To overcome this challenge, we introduce an efficient heuristic strategy to handle high dimensional state space, while also accelerating the convergence behavior. The proposed algorithm is applied to improve two representative trackers, a Siamese-based one and a correlation-filter-based one, to evaluate its generalizability. Their superior performances on several popular benchmarks are clearly demonstrated. Our source code is available at https://github.com/shenjianbing/dqltracking.en_AU
dc.description.sponsorshipThis work was supported in part by the Beijing Natural Science Foundation under Grant 4182056, the CCF-Tencent Open Fund, Zhijiang Lab’s International Talent Fund for Young Professionals, and the Joint Building Program of Beijing Municipal Education Commission. A preliminary version of this work has appeared in CVPR 2018 [1].en_AU
dc.format.mimetypeapplication/pdfen_AU
dc.identifier.issn0162-8828en_AU
dc.identifier.urihttp://hdl.handle.net/1885/317450
dc.language.isoen_AUen_AU
dc.publisherInstitute of Electrical and Electronics Engineers (IEEE Inc)en_AU
dc.rights© 2019 The authorsen_AU
dc.sourceIEEE Transactions on Pattern Analysis and Machine Intelligenceen_AU
dc.subjectHyperparametersen_AU
dc.subjectcontinuous deep q-learningen_AU
dc.subjectreinforcement learningen_AU
dc.subjectvisual object trackingen_AU
dc.titleDynamical hyperparameter optimization via deep reinforcement learning in trackingen_AU
dc.typeJournal articleen_AU
local.bibliographicCitation.issue5en_AU
local.bibliographicCitation.lastpage1529en_AU
local.bibliographicCitation.startpage1515en_AU
local.contributor.affiliationDong, Xingping, Beijing Institute of Technologyen_AU
local.contributor.affiliationShen, Jianbing, Beijing Institute of Technologyen_AU
local.contributor.affiliationWang, Wenguan, Beijing Institute of Technologyen_AU
local.contributor.affiliationShao, Ling, Inception Institute of Artificial Intelligenceen_AU
local.contributor.affiliationLing, Haibin, Stonybrook Universityen_AU
local.contributor.affiliationPorikli, Fatih, College of Engineering, Computing and Cybernetics, ANUen_AU
local.contributor.authoremailu5405232@anu.edu.auen_AU
local.contributor.authoruidPorikli, Fatih, u5405232en_AU
local.description.embargo2099-12-31
local.description.notesImported from ARIESen_AU
local.identifier.absfor400900 - Electronics, sensors and digital hardwareen_AU
local.identifier.ariespublicationa383154xPUB18968en_AU
local.identifier.citationvolume43en_AU
local.identifier.doi10.1109/TPAMI.2019.2956703en_AU
local.identifier.scopusID2-s2.0-85103804043
local.identifier.thomsonIDWOS:000637533800004
local.identifier.uidSubmittedBya383154en_AU
local.publisher.urlhttps://ieeexplore.ieee.org/en_AU
local.type.statusPublished Versionen_AU

Downloads

Original bundle

Now showing 1 - 1 of 1
No Thumbnail Available
Name:
Dynamical_Hyperparameter_Optimization_via_Deep_Reinforcement_Learning_in_Tracking.pdf
Size:
8.12 MB
Format:
Adobe Portable Document Format
Description: