Multi-level Prediction for Overlapped Parcel Segmentation

Zhequan Zhou; Shujing Lyu; Yue Lu

doi:10.1007/978-981-19-2266-4_1

Multi-level Prediction for Overlapped Parcel Segmentation

Zhequan Zhou^*
, Shujing Lyu
, Yue Lu

^*Corresponding author for this work

School of Communication and Electronic Engineering

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

In this paper, we propose a new instance segmentation framework based on a multi-level prediction mechanism, aiming at segmenting overlapped parcels. In this framework, one location on FPN’s feature maps can predict a set of overlapped instances through a multi-level head architecture according to their overlapping order. Besides, to avoid the inherent limit of bounding boxes in object overlapping scenes, our approach is bounding-box free. And we also provide a dataset for overlapped parcel instance segmentation named OLParcel. On a Mask RCNN baseline, our network can improve 4.25% AP, 4.26% recall, and 7.11% MR^{- 2} on our dataset.

Original language	English
Title of host publication	Digital TV and Wireless Multimedia Communications - 18th International Forum, IFTC 2021, Revised Selected Papers
Editors	Guangtao Zhai, Jun Zhou, Hua Yang, Ping An, Xiaokang Yang
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	3-17
Number of pages	15
ISBN (Print)	9789811922657
DOIs	https://doi.org/10.1007/978-981-19-2266-4_1
State	Published - 2022
Event	18th International Forum of Digital Multimedia Communication, IFTC 2021 - Shanghai, China Duration: 3 Dec 2021 → 4 Dec 2021

Publication series

Name	Communications in Computer and Information Science
Volume	1560 CCIS
ISSN (Print)	1865-0929
ISSN (Electronic)	1865-0937

Conference

Conference	18th International Forum of Digital Multimedia Communication, IFTC 2021
Country/Territory	China
City	Shanghai
Period	3/12/21 → 4/12/21

Keywords

Instance segmentation
Overlap
Parcel sorting

Access to Document

10.1007/978-981-19-2266-4_1

Cite this

Zhou, Z., Lyu, S., & Lu, Y. (2022). Multi-level Prediction for Overlapped Parcel Segmentation. In G. Zhai, J. Zhou, H. Yang, P. An, & X. Yang (Eds.), Digital TV and Wireless Multimedia Communications - 18th International Forum, IFTC 2021, Revised Selected Papers (pp. 3-17). (Communications in Computer and Information Science; Vol. 1560 CCIS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-981-19-2266-4_1

Zhou, Zhequan ; Lyu, Shujing ; Lu, Yue. / Multi-level Prediction for Overlapped Parcel Segmentation. Digital TV and Wireless Multimedia Communications - 18th International Forum, IFTC 2021, Revised Selected Papers. editor / Guangtao Zhai ; Jun Zhou ; Hua Yang ; Ping An ; Xiaokang Yang. Springer Science and Business Media Deutschland GmbH, 2022. pp. 3-17 (Communications in Computer and Information Science).

@inproceedings{8314507477b04e9ebd4cf17c5e4a0684,

title = "Multi-level Prediction for Overlapped Parcel Segmentation",

abstract = "In this paper, we propose a new instance segmentation framework based on a multi-level prediction mechanism, aiming at segmenting overlapped parcels. In this framework, one location on FPN{\textquoteright}s feature maps can predict a set of overlapped instances through a multi-level head architecture according to their overlapping order. Besides, to avoid the inherent limit of bounding boxes in object overlapping scenes, our approach is bounding-box free. And we also provide a dataset for overlapped parcel instance segmentation named OLParcel. On a Mask RCNN baseline, our network can improve 4.25\% AP, 4.26\% recall, and 7.11\% MR- 2 on our dataset.",

keywords = "Instance segmentation, Overlap, Parcel sorting",

author = "Zhequan Zhou and Shujing Lyu and Yue Lu",

note = "Publisher Copyright: {\textcopyright} 2022, Springer Nature Singapore Pte Ltd.; 18th International Forum of Digital Multimedia Communication, IFTC 2021 ; Conference date: 03-12-2021 Through 04-12-2021",

year = "2022",

doi = "10.1007/978-981-19-2266-4\_1",

language = "英语",

isbn = "9789811922657",

series = "Communications in Computer and Information Science",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "3--17",

editor = "Guangtao Zhai and Jun Zhou and Hua Yang and Ping An and Xiaokang Yang",

booktitle = "Digital TV and Wireless Multimedia Communications - 18th International Forum, IFTC 2021, Revised Selected Papers",

address = "德国",

}

Zhou, Z, Lyu, S & Lu, Y 2022, Multi-level Prediction for Overlapped Parcel Segmentation. in G Zhai, J Zhou, H Yang, P An & X Yang (eds), Digital TV and Wireless Multimedia Communications - 18th International Forum, IFTC 2021, Revised Selected Papers. Communications in Computer and Information Science, vol. 1560 CCIS, Springer Science and Business Media Deutschland GmbH, pp. 3-17, 18th International Forum of Digital Multimedia Communication, IFTC 2021, Shanghai, China, 3/12/21. https://doi.org/10.1007/978-981-19-2266-4_1

Multi-level Prediction for Overlapped Parcel Segmentation. / Zhou, Zhequan; Lyu, Shujing; Lu, Yue.
Digital TV and Wireless Multimedia Communications - 18th International Forum, IFTC 2021, Revised Selected Papers. ed. / Guangtao Zhai; Jun Zhou; Hua Yang; Ping An; Xiaokang Yang. Springer Science and Business Media Deutschland GmbH, 2022. p. 3-17 (Communications in Computer and Information Science; Vol. 1560 CCIS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Multi-level Prediction for Overlapped Parcel Segmentation

AU - Zhou, Zhequan

AU - Lyu, Shujing

AU - Lu, Yue

PY - 2022

Y1 - 2022

N2 - In this paper, we propose a new instance segmentation framework based on a multi-level prediction mechanism, aiming at segmenting overlapped parcels. In this framework, one location on FPN’s feature maps can predict a set of overlapped instances through a multi-level head architecture according to their overlapping order. Besides, to avoid the inherent limit of bounding boxes in object overlapping scenes, our approach is bounding-box free. And we also provide a dataset for overlapped parcel instance segmentation named OLParcel. On a Mask RCNN baseline, our network can improve 4.25% AP, 4.26% recall, and 7.11% MR- 2 on our dataset.

AB - In this paper, we propose a new instance segmentation framework based on a multi-level prediction mechanism, aiming at segmenting overlapped parcels. In this framework, one location on FPN’s feature maps can predict a set of overlapped instances through a multi-level head architecture according to their overlapping order. Besides, to avoid the inherent limit of bounding boxes in object overlapping scenes, our approach is bounding-box free. And we also provide a dataset for overlapped parcel instance segmentation named OLParcel. On a Mask RCNN baseline, our network can improve 4.25% AP, 4.26% recall, and 7.11% MR- 2 on our dataset.

KW - Instance segmentation

KW - Overlap

KW - Parcel sorting

UR - https://www.scopus.com/pages/publications/85128906981

U2 - 10.1007/978-981-19-2266-4_1

DO - 10.1007/978-981-19-2266-4_1

M3 - 会议稿件

AN - SCOPUS:85128906981

SN - 9789811922657

T3 - Communications in Computer and Information Science

SP - 3

EP - 17

BT - Digital TV and Wireless Multimedia Communications - 18th International Forum, IFTC 2021, Revised Selected Papers

A2 - Zhai, Guangtao

A2 - Zhou, Jun

A2 - Yang, Hua

A2 - An, Ping

A2 - Yang, Xiaokang

PB - Springer Science and Business Media Deutschland GmbH

T2 - 18th International Forum of Digital Multimedia Communication, IFTC 2021

Y2 - 3 December 2021 through 4 December 2021

ER -

Zhou Z, Lyu S, Lu Y. Multi-level Prediction for Overlapped Parcel Segmentation. In Zhai G, Zhou J, Yang H, An P, Yang X, editors, Digital TV and Wireless Multimedia Communications - 18th International Forum, IFTC 2021, Revised Selected Papers. Springer Science and Business Media Deutschland GmbH. 2022. p. 3-17. (Communications in Computer and Information Science). doi: 10.1007/978-981-19-2266-4_1

Multi-level Prediction for Overlapped Parcel Segmentation

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this