Learning Snippet Relatedness Based on LSTM for Temporal Action Proposal Generation

Hyunjun Eun; Jinyoung Moon; Jongyoul Park; Chanho Jung; Changick Kim

doi:10.7840/kics.2020.45.6.975

Learning Snippet Relatedness Based on LSTM for Temporal Action Proposal Generation

Hyunjun Eun, Jinyoung Moon, Jongyoul Park, Chanho Jung, Changick Kim

Dept. of Applied Artificial Intelligence

Research output: Contribution to journal › Article › peer-review

Abstract

Recent temporal action proposal generation approaches are based on temporal convolutional networks. In this paper, different from this, we propose to use LSTM for sequential modeling on actions. The propose method based on LSTM evaluates snippet relatedness to define temporal action intervals. Snippet relatedness indicates which snippets are included in the same action instance. By conducting experiments on the THUMOS-14 dataset, we demonstrate the superiority of the proposed method. We also analyze our method in diverse aspects.

Original language	English
Pages (from-to)	975-978
Number of pages	4
Journal	Journal of Korean Institute of Communications and Information Sciences
Volume	45
Issue number	6
DOIs	https://doi.org/10.7840/kics.2020.45.6.975
State	Published - Jun 2020

Keywords

LSTM
snippet relatedness
temporal action proposal generation

Access to Document

10.7840/kics.2020.45.6.975

Cite this

@article{52bce767e8034e629974703917bd6f42,

title = "Learning Snippet Relatedness Based on LSTM for Temporal Action Proposal Generation",

abstract = "Recent temporal action proposal generation approaches are based on temporal convolutional networks. In this paper, different from this, we propose to use LSTM for sequential modeling on actions. The propose method based on LSTM evaluates snippet relatedness to define temporal action intervals. Snippet relatedness indicates which snippets are included in the same action instance. By conducting experiments on the THUMOS-14 dataset, we demonstrate the superiority of the proposed method. We also analyze our method in diverse aspects.",

keywords = "LSTM, snippet relatedness, temporal action proposal generation",

author = "Hyunjun Eun and Jinyoung Moon and Jongyoul Park and Chanho Jung and Changick Kim",

year = "2020",

month = jun,

doi = "10.7840/kics.2020.45.6.975",

language = "English",

volume = "45",

pages = "975--978",

journal = "Journal of Korean Institute of Communications and Information Sciences",

issn = "1226-4717",

number = "6",

}

TY - JOUR

T1 - Learning Snippet Relatedness Based on LSTM for Temporal Action Proposal Generation

AU - Eun, Hyunjun

AU - Moon, Jinyoung

AU - Park, Jongyoul

AU - Jung, Chanho

AU - Kim, Changick

PY - 2020/6

Y1 - 2020/6

N2 - Recent temporal action proposal generation approaches are based on temporal convolutional networks. In this paper, different from this, we propose to use LSTM for sequential modeling on actions. The propose method based on LSTM evaluates snippet relatedness to define temporal action intervals. Snippet relatedness indicates which snippets are included in the same action instance. By conducting experiments on the THUMOS-14 dataset, we demonstrate the superiority of the proposed method. We also analyze our method in diverse aspects.

AB - Recent temporal action proposal generation approaches are based on temporal convolutional networks. In this paper, different from this, we propose to use LSTM for sequential modeling on actions. The propose method based on LSTM evaluates snippet relatedness to define temporal action intervals. Snippet relatedness indicates which snippets are included in the same action instance. By conducting experiments on the THUMOS-14 dataset, we demonstrate the superiority of the proposed method. We also analyze our method in diverse aspects.

KW - LSTM

KW - snippet relatedness

KW - temporal action proposal generation

UR - https://www.scopus.com/pages/publications/85189564035

U2 - 10.7840/kics.2020.45.6.975

DO - 10.7840/kics.2020.45.6.975

M3 - Article

AN - SCOPUS:85189564035

SN - 1226-4717

VL - 45

SP - 975

EP - 978

JO - Journal of Korean Institute of Communications and Information Sciences

JF - Journal of Korean Institute of Communications and Information Sciences

IS - 6

ER -

Learning Snippet Relatedness Based on LSTM for Temporal Action Proposal Generation

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this