New Branch Optimization Design Based on RefineNet

Gengyun Ren; Xiujie Qu

doi:10.1109/ICAICA50127.2020.9182507

New Branch Optimization Design Based on RefineNet

Gengyun Ren, Xiujie Qu

School of Integrated Circuits and Electronics

Beijing Institute of Technology

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

Abstract

Usually, in semantic segmentation, researchers only use the final output layer for the network training, which means, we ignore information from other layers. To this situation, we design new branches for output, which can output rough small-scale prediction from the middle of the network. With the new branches, the middle of the network will get better control and guidance in the process of training, and also we can make better use of balanced semantic information and spatial information in the middle of the network. With the addition of new branches, more information can be used in the network training. This makes mIoU accuracy 4% increase on CamVid dataset based on RefineNet.

Original language	English
Title of host publication	Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	591-594
Number of pages	4
ISBN (Electronic)	9781728170046
DOIs	https://doi.org/10.1109/ICAICA50127.2020.9182507
Publication status	Published - Jun 2020
Event	2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020 - Dalian, China Duration: 27 Jun 2020 → 29 Jun 2020

Publication series

Name	Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020

Conference

Conference	2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020
Country/Territory	China
City	Dalian
Period	27/06/20 → 29/06/20

Keywords

Convolutional neural network
RefineNet
computer vision
semantic segmentation

Access to Document

10.1109/ICAICA50127.2020.9182507

Cite this

Ren, G., & Qu, X. (2020). New Branch Optimization Design Based on RefineNet. In Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020 (pp. 591-594). Article 9182507 (Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICAICA50127.2020.9182507

Ren, Gengyun ; Qu, Xiujie. / New Branch Optimization Design Based on RefineNet. Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020. Institute of Electrical and Electronics Engineers Inc., 2020. pp. 591-594 (Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020).

@inproceedings{f434c48c27404d23be48e801c1f1ded8,

title = "New Branch Optimization Design Based on RefineNet",

abstract = "Usually, in semantic segmentation, researchers only use the final output layer for the network training, which means, we ignore information from other layers. To this situation, we design new branches for output, which can output rough small-scale prediction from the middle of the network. With the new branches, the middle of the network will get better control and guidance in the process of training, and also we can make better use of balanced semantic information and spatial information in the middle of the network. With the addition of new branches, more information can be used in the network training. This makes mIoU accuracy 4% increase on CamVid dataset based on RefineNet.",

keywords = "Convolutional neural network, RefineNet, computer vision, semantic segmentation",

author = "Gengyun Ren and Xiujie Qu",

note = "Publisher Copyright: {\textcopyright} 2020 IEEE.; 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020 ; Conference date: 27-06-2020 Through 29-06-2020",

year = "2020",

month = jun,

doi = "10.1109/ICAICA50127.2020.9182507",

language = "English",

series = "Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "591--594",

booktitle = "Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020",

address = "United States",

}

Ren, G & Qu, X 2020, New Branch Optimization Design Based on RefineNet. in Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020., 9182507, Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020, Institute of Electrical and Electronics Engineers Inc., pp. 591-594, 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020, Dalian, China, 27/06/20. https://doi.org/10.1109/ICAICA50127.2020.9182507

New Branch Optimization Design Based on RefineNet. / Ren, Gengyun; Qu, Xiujie.
Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020. Institute of Electrical and Electronics Engineers Inc., 2020. p. 591-594 9182507 (Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020).

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

TY - GEN

T1 - New Branch Optimization Design Based on RefineNet

AU - Ren, Gengyun

AU - Qu, Xiujie

PY - 2020/6

Y1 - 2020/6

N2 - Usually, in semantic segmentation, researchers only use the final output layer for the network training, which means, we ignore information from other layers. To this situation, we design new branches for output, which can output rough small-scale prediction from the middle of the network. With the new branches, the middle of the network will get better control and guidance in the process of training, and also we can make better use of balanced semantic information and spatial information in the middle of the network. With the addition of new branches, more information can be used in the network training. This makes mIoU accuracy 4% increase on CamVid dataset based on RefineNet.

AB - Usually, in semantic segmentation, researchers only use the final output layer for the network training, which means, we ignore information from other layers. To this situation, we design new branches for output, which can output rough small-scale prediction from the middle of the network. With the new branches, the middle of the network will get better control and guidance in the process of training, and also we can make better use of balanced semantic information and spatial information in the middle of the network. With the addition of new branches, more information can be used in the network training. This makes mIoU accuracy 4% increase on CamVid dataset based on RefineNet.

KW - Convolutional neural network

KW - RefineNet

KW - computer vision

KW - semantic segmentation

UR - http://www.scopus.com/inward/record.url?scp=85092179842&partnerID=8YFLogxK

U2 - 10.1109/ICAICA50127.2020.9182507

DO - 10.1109/ICAICA50127.2020.9182507

M3 - Conference contribution

AN - SCOPUS:85092179842

T3 - Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020

SP - 591

EP - 594

BT - Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020

Y2 - 27 June 2020 through 29 June 2020

ER -

Ren G, Qu X. New Branch Optimization Design Based on RefineNet. In Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020. Institute of Electrical and Electronics Engineers Inc. 2020. p. 591-594. 9182507. (Proceedings of 2020 IEEE International Conference on Artificial Intelligence and Computer Applications, ICAICA 2020). doi: 10.1109/ICAICA50127.2020.9182507

New Branch Optimization Design Based on RefineNet

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this