Hybrid connection network for semantic segmentation

Xiao Liang, Seiichiro Kamata

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In recent years, deep convolutional neural networks like ResNet and DenseNet with short cut connected to each layer can be more accurate and easier to train. Although deep convolutional neural networks show their strength in many computer vision tasks, there is still a challenge to get more precise per-pixel prediction for semantic image segmentation task via deep convolutional neural networks. In this paper, we propose a hybrid connection network architecture for semantic segmentation which consists of an encoder network for extracting different scale feature maps and a decoder network for recovering extracted feature maps with the resolution of the input image. This architecture includes several skip connection paths between encoder and decoder. The paths help to fuse both localization information and global information. We show that our architecture can be quickly trained end-to-end without pre-training on an additional dataset and performs comparable results on semantic segmentation benchmark datasets such as PASCAL VOC 2012.

Original languageEnglish
Title of host publicationTenth International Conference on Digital Image Processing, ICDIP 2018
EditorsJenq-Neng Hwang, Xudong Jiang
PublisherSPIE
Volume10806
ISBN (Print)9781510621992
DOIs
Publication statusPublished - 2018 Jan 1
Event10th International Conference on Digital Image Processing, ICDIP 2018 - Shanghai, China
Duration: 2018 May 112018 May 14

Other

Other10th International Conference on Digital Image Processing, ICDIP 2018
CountryChina
CityShanghai
Period18/5/1118/5/14

Fingerprint

semantics
Segmentation
Semantics
Neural Networks
Encoder
Neural networks
Path
decoders
Electric fuses
Network Architecture
coders
Network architecture
Volatile organic compounds
Image segmentation
Image Segmentation
Computer Vision
Computer vision
Pixel
Pixels
Benchmark

Keywords

  • Computer vision
  • DenseNet
  • Fully convolutional network
  • ResNet
  • Semantic image segmentation

ASJC Scopus subject areas

  • Electronic, Optical and Magnetic Materials
  • Condensed Matter Physics
  • Computer Science Applications
  • Applied Mathematics
  • Electrical and Electronic Engineering

Cite this

Liang, X., & Kamata, S. (2018). Hybrid connection network for semantic segmentation. In J-N. Hwang, & X. Jiang (Eds.), Tenth International Conference on Digital Image Processing, ICDIP 2018 (Vol. 10806). [108066P] SPIE. https://doi.org/10.1117/12.2502963

Hybrid connection network for semantic segmentation. / Liang, Xiao; Kamata, Seiichiro.

Tenth International Conference on Digital Image Processing, ICDIP 2018. ed. / Jenq-Neng Hwang; Xudong Jiang. Vol. 10806 SPIE, 2018. 108066P.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Liang, X & Kamata, S 2018, Hybrid connection network for semantic segmentation. in J-N Hwang & X Jiang (eds), Tenth International Conference on Digital Image Processing, ICDIP 2018. vol. 10806, 108066P, SPIE, 10th International Conference on Digital Image Processing, ICDIP 2018, Shanghai, China, 18/5/11. https://doi.org/10.1117/12.2502963
Liang X, Kamata S. Hybrid connection network for semantic segmentation. In Hwang J-N, Jiang X, editors, Tenth International Conference on Digital Image Processing, ICDIP 2018. Vol. 10806. SPIE. 2018. 108066P https://doi.org/10.1117/12.2502963
Liang, Xiao ; Kamata, Seiichiro. / Hybrid connection network for semantic segmentation. Tenth International Conference on Digital Image Processing, ICDIP 2018. editor / Jenq-Neng Hwang ; Xudong Jiang. Vol. 10806 SPIE, 2018.
@inproceedings{5f63653ed92c4fbda2d78c13b565a7ed,
title = "Hybrid connection network for semantic segmentation",
abstract = "In recent years, deep convolutional neural networks like ResNet and DenseNet with short cut connected to each layer can be more accurate and easier to train. Although deep convolutional neural networks show their strength in many computer vision tasks, there is still a challenge to get more precise per-pixel prediction for semantic image segmentation task via deep convolutional neural networks. In this paper, we propose a hybrid connection network architecture for semantic segmentation which consists of an encoder network for extracting different scale feature maps and a decoder network for recovering extracted feature maps with the resolution of the input image. This architecture includes several skip connection paths between encoder and decoder. The paths help to fuse both localization information and global information. We show that our architecture can be quickly trained end-to-end without pre-training on an additional dataset and performs comparable results on semantic segmentation benchmark datasets such as PASCAL VOC 2012.",
keywords = "Computer vision, DenseNet, Fully convolutional network, ResNet, Semantic image segmentation",
author = "Xiao Liang and Seiichiro Kamata",
year = "2018",
month = "1",
day = "1",
doi = "10.1117/12.2502963",
language = "English",
isbn = "9781510621992",
volume = "10806",
editor = "Jenq-Neng Hwang and Xudong Jiang",
booktitle = "Tenth International Conference on Digital Image Processing, ICDIP 2018",
publisher = "SPIE",

}

TY - GEN

T1 - Hybrid connection network for semantic segmentation

AU - Liang, Xiao

AU - Kamata, Seiichiro

PY - 2018/1/1

Y1 - 2018/1/1

N2 - In recent years, deep convolutional neural networks like ResNet and DenseNet with short cut connected to each layer can be more accurate and easier to train. Although deep convolutional neural networks show their strength in many computer vision tasks, there is still a challenge to get more precise per-pixel prediction for semantic image segmentation task via deep convolutional neural networks. In this paper, we propose a hybrid connection network architecture for semantic segmentation which consists of an encoder network for extracting different scale feature maps and a decoder network for recovering extracted feature maps with the resolution of the input image. This architecture includes several skip connection paths between encoder and decoder. The paths help to fuse both localization information and global information. We show that our architecture can be quickly trained end-to-end without pre-training on an additional dataset and performs comparable results on semantic segmentation benchmark datasets such as PASCAL VOC 2012.

AB - In recent years, deep convolutional neural networks like ResNet and DenseNet with short cut connected to each layer can be more accurate and easier to train. Although deep convolutional neural networks show their strength in many computer vision tasks, there is still a challenge to get more precise per-pixel prediction for semantic image segmentation task via deep convolutional neural networks. In this paper, we propose a hybrid connection network architecture for semantic segmentation which consists of an encoder network for extracting different scale feature maps and a decoder network for recovering extracted feature maps with the resolution of the input image. This architecture includes several skip connection paths between encoder and decoder. The paths help to fuse both localization information and global information. We show that our architecture can be quickly trained end-to-end without pre-training on an additional dataset and performs comparable results on semantic segmentation benchmark datasets such as PASCAL VOC 2012.

KW - Computer vision

KW - DenseNet

KW - Fully convolutional network

KW - ResNet

KW - Semantic image segmentation

UR - http://www.scopus.com/inward/record.url?scp=85052613736&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85052613736&partnerID=8YFLogxK

U2 - 10.1117/12.2502963

DO - 10.1117/12.2502963

M3 - Conference contribution

AN - SCOPUS:85052613736

SN - 9781510621992

VL - 10806

BT - Tenth International Conference on Digital Image Processing, ICDIP 2018

A2 - Hwang, Jenq-Neng

A2 - Jiang, Xudong

PB - SPIE

ER -