SegFast-V2: Semantic image segmentation with less parameters in deep learning for autonomous driving

Swarnendu Ghosh, Anisha Pal, Shourya Jaiswal, K. C. Santosh, Nibaran Das, Mita Nasipuri

Research output: Contribution to journalArticle

Abstract

Semantic image segmentation can be used in various driving applications, such as automatic braking, road sign alerts, park assists, and pedestrian warnings. More often, AI applications, such as autonomous modules are available in expensive vehicles. It would be appreciated if such facilities can be made available in the lower end of the price spectrum. Existing methodologies, come with a costly overhead with large number of parameters and need of costly hardware. Within this scope, the key contribution of this work is to promote the possibility of compact semantic image segmentation so that it can be extended to deploy AI based solutions to less expensive vehicles. While developing cheap and fast models one must also not compromise the factor of reliability and robustness. The proposed work is primarily based on our previous model named “SegFast”, and is aimed to perform thorough analysis across a multitude of datasets. Beside “spark” modules and depth-wise separable transposed convolutions, kernel factorization is implemented to further reduce the number of parameters. The effect of MobileNet as an encoder to our model has also been analyzed. The proposed method shows a promising decrease in the number of parameters and significant gain in terms of runtime even on a single CPU environment. Despite all those speedups, the proposed approach performs at a similar level to many popular but heavier networks, such as SegNet, UNet, PSPNet, and FCN.

Original languageEnglish
Pages (from-to)3145-3154
Number of pages10
JournalInternational Journal of Machine Learning and Cybernetics
Volume10
Issue number11
DOIs
Publication statusPublished - 01-11-2019
Externally publishedYes

Fingerprint

Image segmentation
Semantics
Braking
Factorization
Electric sparks
Convolution
Program processors
Hardware
Deep learning

All Science Journal Classification (ASJC) codes

  • Software
  • Computer Vision and Pattern Recognition
  • Artificial Intelligence

Cite this

Ghosh, Swarnendu ; Pal, Anisha ; Jaiswal, Shourya ; Santosh, K. C. ; Das, Nibaran ; Nasipuri, Mita. / SegFast-V2 : Semantic image segmentation with less parameters in deep learning for autonomous driving. In: International Journal of Machine Learning and Cybernetics. 2019 ; Vol. 10, No. 11. pp. 3145-3154.
@article{8b5cfef1c8c044f1a5821a75499a4b41,
title = "SegFast-V2: Semantic image segmentation with less parameters in deep learning for autonomous driving",
abstract = "Semantic image segmentation can be used in various driving applications, such as automatic braking, road sign alerts, park assists, and pedestrian warnings. More often, AI applications, such as autonomous modules are available in expensive vehicles. It would be appreciated if such facilities can be made available in the lower end of the price spectrum. Existing methodologies, come with a costly overhead with large number of parameters and need of costly hardware. Within this scope, the key contribution of this work is to promote the possibility of compact semantic image segmentation so that it can be extended to deploy AI based solutions to less expensive vehicles. While developing cheap and fast models one must also not compromise the factor of reliability and robustness. The proposed work is primarily based on our previous model named “SegFast”, and is aimed to perform thorough analysis across a multitude of datasets. Beside “spark” modules and depth-wise separable transposed convolutions, kernel factorization is implemented to further reduce the number of parameters. The effect of MobileNet as an encoder to our model has also been analyzed. The proposed method shows a promising decrease in the number of parameters and significant gain in terms of runtime even on a single CPU environment. Despite all those speedups, the proposed approach performs at a similar level to many popular but heavier networks, such as SegNet, UNet, PSPNet, and FCN.",
author = "Swarnendu Ghosh and Anisha Pal and Shourya Jaiswal and Santosh, {K. C.} and Nibaran Das and Mita Nasipuri",
year = "2019",
month = "11",
day = "1",
doi = "10.1007/s13042-019-01005-5",
language = "English",
volume = "10",
pages = "3145--3154",
journal = "International Journal of Machine Learning and Cybernetics",
issn = "1868-8071",
publisher = "Springer Science + Business Media",
number = "11",

}

SegFast-V2 : Semantic image segmentation with less parameters in deep learning for autonomous driving. / Ghosh, Swarnendu; Pal, Anisha; Jaiswal, Shourya; Santosh, K. C.; Das, Nibaran; Nasipuri, Mita.

In: International Journal of Machine Learning and Cybernetics, Vol. 10, No. 11, 01.11.2019, p. 3145-3154.

Research output: Contribution to journalArticle

TY - JOUR

T1 - SegFast-V2

T2 - Semantic image segmentation with less parameters in deep learning for autonomous driving

AU - Ghosh, Swarnendu

AU - Pal, Anisha

AU - Jaiswal, Shourya

AU - Santosh, K. C.

AU - Das, Nibaran

AU - Nasipuri, Mita

PY - 2019/11/1

Y1 - 2019/11/1

N2 - Semantic image segmentation can be used in various driving applications, such as automatic braking, road sign alerts, park assists, and pedestrian warnings. More often, AI applications, such as autonomous modules are available in expensive vehicles. It would be appreciated if such facilities can be made available in the lower end of the price spectrum. Existing methodologies, come with a costly overhead with large number of parameters and need of costly hardware. Within this scope, the key contribution of this work is to promote the possibility of compact semantic image segmentation so that it can be extended to deploy AI based solutions to less expensive vehicles. While developing cheap and fast models one must also not compromise the factor of reliability and robustness. The proposed work is primarily based on our previous model named “SegFast”, and is aimed to perform thorough analysis across a multitude of datasets. Beside “spark” modules and depth-wise separable transposed convolutions, kernel factorization is implemented to further reduce the number of parameters. The effect of MobileNet as an encoder to our model has also been analyzed. The proposed method shows a promising decrease in the number of parameters and significant gain in terms of runtime even on a single CPU environment. Despite all those speedups, the proposed approach performs at a similar level to many popular but heavier networks, such as SegNet, UNet, PSPNet, and FCN.

AB - Semantic image segmentation can be used in various driving applications, such as automatic braking, road sign alerts, park assists, and pedestrian warnings. More often, AI applications, such as autonomous modules are available in expensive vehicles. It would be appreciated if such facilities can be made available in the lower end of the price spectrum. Existing methodologies, come with a costly overhead with large number of parameters and need of costly hardware. Within this scope, the key contribution of this work is to promote the possibility of compact semantic image segmentation so that it can be extended to deploy AI based solutions to less expensive vehicles. While developing cheap and fast models one must also not compromise the factor of reliability and robustness. The proposed work is primarily based on our previous model named “SegFast”, and is aimed to perform thorough analysis across a multitude of datasets. Beside “spark” modules and depth-wise separable transposed convolutions, kernel factorization is implemented to further reduce the number of parameters. The effect of MobileNet as an encoder to our model has also been analyzed. The proposed method shows a promising decrease in the number of parameters and significant gain in terms of runtime even on a single CPU environment. Despite all those speedups, the proposed approach performs at a similar level to many popular but heavier networks, such as SegNet, UNet, PSPNet, and FCN.

UR - http://www.scopus.com/inward/record.url?scp=85072036415&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85072036415&partnerID=8YFLogxK

U2 - 10.1007/s13042-019-01005-5

DO - 10.1007/s13042-019-01005-5

M3 - Article

AN - SCOPUS:85072036415

VL - 10

SP - 3145

EP - 3154

JO - International Journal of Machine Learning and Cybernetics

JF - International Journal of Machine Learning and Cybernetics

SN - 1868-8071

IS - 11

ER -