닫기
Article Search
닫기

Original Article

Split Viewer

International Journal of Fuzzy Logic and Intelligent Systems 2021; 21(4): 401-408

Published online December 25, 2021

https://doi.org/10.5391/IJFIS.2021.21.4.401

© The Korean Institute of Intelligent Systems

MPFANet: Semantic Segmentation Using Multiple Path Feature Aggregation

Wang-Su Jeon1 and Sang-Yong Rhee2

1Department of IT Convergence Engineering, Kyungnam University, Changwon, Korea
2Department of Computer Engineering, Kyungnam University, Changwon, Korea

Correspondence to :
Sang-Yong Rhee (syrhee@kyungnam.ac.kr)
*These authors contributed equally to this work.

Received: November 4, 2020; Revised: September 23, 2021; Accepted: November 8, 2021

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Go to

Image segmentation is the process of simplifying the analysis of the meaning or the front to say the process of dividing the image into a set of multiple pixels. The multiple path feature aggregation (MPFA) method proposed in this paper aims to extract various information of an object, and uses conventional pyramid pooling or the extraction of various sized features. This information can be combined with different regional features to obtain the overall feature information. We split four paths to extract numerous local features, and the results showed that the mean intersection over union (mIOU) is 81.6% for the validation data from the PASCAL VOC 2012 dataset, and a better performance than the existing DeepLab model was demonstrated.

Keywords: MPFA, Semantic segmentation, Feature aggregation, CNN, Inverted residual block, Local context

Acknowledgments

Go to
This work was supported by a National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No. 2020R1F1A1075968).

Conflict of Interest

Go to

No potential conflicts of interest relevant to this article are reported.

Biographies

Go to

Wang-Su Jeon received his B.S. and M. S. degrees in Computer Engineering and IT Convergence Engineering from Kyungnam University, Masan, Korea, in 2016 and 2018, and is currently pursuing the Ph.D. degree in IT Convergence Engineering at Kyungnam University, Masan, Korea. His present interests include computer vision.

E-mail : jws2218@naver.com

Sang-Yong Rhee received his B.S. and M.S. degrees in Industrial Engineering from Korea University, Seoul, Korea, in 1982 and 1984, respectively, and his Ph.D. degree in Industrial Engineering at Pohang University, Pohang, Korea. He is currently a professor at the Computer Engineering, Kyungnam University, Masan, Korea. His research interests include computer vision, augmented reality, deep learning and human-robot interface.

E-mail : syrhee@kyungnam.ac.kr

Article

Original Article

International Journal of Fuzzy Logic and Intelligent Systems 2021; 21(4): 401-408

Published online December 25, 2021 https://doi.org/10.5391/IJFIS.2021.21.4.401

Copyright © The Korean Institute of Intelligent Systems.

MPFANet: Semantic Segmentation Using Multiple Path Feature Aggregation

Wang-Su Jeon1 and Sang-Yong Rhee2

1Department of IT Convergence Engineering, Kyungnam University, Changwon, Korea
2Department of Computer Engineering, Kyungnam University, Changwon, Korea

Correspondence to:Sang-Yong Rhee (syrhee@kyungnam.ac.kr)
*These authors contributed equally to this work.

Received: November 4, 2020; Revised: September 23, 2021; Accepted: November 8, 2021

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Image segmentation is the process of simplifying the analysis of the meaning or the front to say the process of dividing the image into a set of multiple pixels. The multiple path feature aggregation (MPFA) method proposed in this paper aims to extract various information of an object, and uses conventional pyramid pooling or the extraction of various sized features. This information can be combined with different regional features to obtain the overall feature information. We split four paths to extract numerous local features, and the results showed that the mean intersection over union (mIOU) is 81.6% for the validation data from the PASCAL VOC 2012 dataset, and a better performance than the existing DeepLab model was demonstrated.

Keywords: MPFA, Semantic segmentation, Feature aggregation, CNN, Inverted residual block, Local context

Fig 1.

Download
Figure 1.

Type of segmentation: (a) input, (b) instance segmentation, (c) semantic segmentation, and (d) panoptic segmentation. The images are from [1].

The International Journal of Fuzzy Logic and Intelligent Systems 2021; 21: 401-408https://doi.org/10.5391/IJFIS.2021.21.4.401

Fig 2.

Download
Figure 2.

Multiple path feature aggregation network architecture.

The International Journal of Fuzzy Logic and Intelligent Systems 2021; 21: 401-408https://doi.org/10.5391/IJFIS.2021.21.4.401

Fig 3.

Download
Figure 3.

Atrous spatial pyramid pooling structure.

The International Journal of Fuzzy Logic and Intelligent Systems 2021; 21: 401-408https://doi.org/10.5391/IJFIS.2021.21.4.401

Fig 4.

Download
Figure 4.

Multi-scale residual feature structure.

The International Journal of Fuzzy Logic and Intelligent Systems 2021; 21: 401-408https://doi.org/10.5391/IJFIS.2021.21.4.401

Fig 5.

Download
Figure 5.

Linear bottleneck structure: (a) residual block and (b) inverted residual block.

The International Journal of Fuzzy Logic and Intelligent Systems 2021; 21: 401-408https://doi.org/10.5391/IJFIS.2021.21.4.401

Fig 6.

Download
Figure 6.

Pyramid pooling module structure.

The International Journal of Fuzzy Logic and Intelligent Systems 2021; 21: 401-408https://doi.org/10.5391/IJFIS.2021.21.4.401

Fig 7.

Download
Figure 7.

Visual measurement of PASCAL VOC 2012 data. Results of the proposed method are compared with the baseline.

The International Journal of Fuzzy Logic and Intelligent Systems 2021; 21: 401-408https://doi.org/10.5391/IJFIS.2021.21.4.401

Fig 8.

Download
Figure 8.

Visual measurement of PASCAL VOC 2012 data. The results of the proposed method are compared with the baseline.

The International Journal of Fuzzy Logic and Intelligent Systems 2021; 21: 401-408https://doi.org/10.5391/IJFIS.2021.21.4.401

Table 1 . Mean IOU and fps measurement according to MPFA size.

MethodMean IOU(%)fps
ResNet50-MPFA51281.612
ResNet50-MPFA25678.324
ResNet50-MPFA12875.532
ResNet50-MPFA6472.340

Table 2 . Mean IOU measurement according to batch size and iterations.

MethodBatch sizeIterationMean IOU(%)
ResNet50-MPFA512410k77.0
ResNet50-MPFA512420k77.7
ResNet50-MPFA5121020k78.3
ResNet50-MPFA5121020k81.6

Table 3 . PASCAL-VOC-2012 validation data of each model compared with mean IOU.

MethodMean IOU(%)
ResNet-GCN [36]81.0
DFN [35]80.6
DeepLab v3+ [34]79.3
FastDenseNas-arch0 [33]78.0
DeepLab-CRF [16]77.69
Proposed method81.6

Article Tools

Metrics

Share this article on :

Related articles in IJFIS