Training of convolutional networks on multiple heterogeneous datasets for street scene semantic segmentation

Onderzoeksoutput: Hoofdstuk in Boek/Rapport/CongresprocedureConferentiebijdrageAcademicpeer review

4 Citaties (Scopus)
1 Downloads (Pure)

Uittreksel

We propose a convolutional network with hierarchical classifiers for per-pixel semantic segmentation, which is able to be trained on multiple, heterogeneous datasets and exploit their semantic hierarchy. Our network is the first to be simultaneously trained on three different datasets from the intelligent vehicles domain, i.e. Cityscapes, GTSDB and Mapillary Vistas, and is able to handle different semantic levelof-detail, class imbalances, and different annotation types, i.e. dense per-pixel and sparse bounding-box labels. We assess our hierarchical approach, by comparing against flat, nonhierarchical classifiers and we show improvements in mean pixel accuracy of 13.0% for Cityscapes classes and 2.4% for Vistas classes and 32.3% for GTSDB classes. Our implementation achieves inference rates of 17 fps at a resolution of 520 x 706 for 108 classes running on a GPU.

Originele taal-2Engels
Titel2018 IEEE Intelligent Vehicles Symposium, IV 2018
UitgeverijInstitute of Electrical and Electronics Engineers
Pagina's1045-1050
Aantal pagina's6
ISBN van elektronische versie9781538644522
DOI's
StatusGepubliceerd - 22 okt 2018
Evenement2018 IEEE Intelligent Vehicles Symposium, IV 2018 - Changshu, Suzhou, China
Duur: 26 sep 201830 sep 2018

Congres

Congres2018 IEEE Intelligent Vehicles Symposium, IV 2018
LandChina
StadChangshu, Suzhou
Periode26/09/1830/09/18

Vingerafdruk

Segmentation
Pixels
Semantics
Classifiers
Pixel
Intelligent vehicle highway systems
Classifier
Labels
Intelligent Vehicle
Annotation
Training
Class
Graphics processing unit

Citeer dit

Meletis, P., & Dubbelman, G. (2018). Training of convolutional networks on multiple heterogeneous datasets for street scene semantic segmentation. In 2018 IEEE Intelligent Vehicles Symposium, IV 2018 (blz. 1045-1050). [8500398] Institute of Electrical and Electronics Engineers. https://doi.org/10.1109/IVS.2018.8500398
Meletis, Panagiotis ; Dubbelman, Gijs. / Training of convolutional networks on multiple heterogeneous datasets for street scene semantic segmentation. 2018 IEEE Intelligent Vehicles Symposium, IV 2018. Institute of Electrical and Electronics Engineers, 2018. blz. 1045-1050
@inproceedings{8faf5dc5ba604b72a78c31bcf422b42e,
title = "Training of convolutional networks on multiple heterogeneous datasets for street scene semantic segmentation",
abstract = "We propose a convolutional network with hierarchical classifiers for per-pixel semantic segmentation, which is able to be trained on multiple, heterogeneous datasets and exploit their semantic hierarchy. Our network is the first to be simultaneously trained on three different datasets from the intelligent vehicles domain, i.e. Cityscapes, GTSDB and Mapillary Vistas, and is able to handle different semantic levelof-detail, class imbalances, and different annotation types, i.e. dense per-pixel and sparse bounding-box labels. We assess our hierarchical approach, by comparing against flat, nonhierarchical classifiers and we show improvements in mean pixel accuracy of 13.0{\%} for Cityscapes classes and 2.4{\%} for Vistas classes and 32.3{\%} for GTSDB classes. Our implementation achieves inference rates of 17 fps at a resolution of 520 x 706 for 108 classes running on a GPU.",
author = "Panagiotis Meletis and Gijs Dubbelman",
year = "2018",
month = "10",
day = "22",
doi = "10.1109/IVS.2018.8500398",
language = "English",
pages = "1045--1050",
booktitle = "2018 IEEE Intelligent Vehicles Symposium, IV 2018",
publisher = "Institute of Electrical and Electronics Engineers",
address = "United States",

}

Meletis, P & Dubbelman, G 2018, Training of convolutional networks on multiple heterogeneous datasets for street scene semantic segmentation. in 2018 IEEE Intelligent Vehicles Symposium, IV 2018., 8500398, Institute of Electrical and Electronics Engineers, blz. 1045-1050, Changshu, Suzhou, China, 26/09/18. https://doi.org/10.1109/IVS.2018.8500398

Training of convolutional networks on multiple heterogeneous datasets for street scene semantic segmentation. / Meletis, Panagiotis; Dubbelman, Gijs.

2018 IEEE Intelligent Vehicles Symposium, IV 2018. Institute of Electrical and Electronics Engineers, 2018. blz. 1045-1050 8500398.

Onderzoeksoutput: Hoofdstuk in Boek/Rapport/CongresprocedureConferentiebijdrageAcademicpeer review

TY - GEN

T1 - Training of convolutional networks on multiple heterogeneous datasets for street scene semantic segmentation

AU - Meletis, Panagiotis

AU - Dubbelman, Gijs

PY - 2018/10/22

Y1 - 2018/10/22

N2 - We propose a convolutional network with hierarchical classifiers for per-pixel semantic segmentation, which is able to be trained on multiple, heterogeneous datasets and exploit their semantic hierarchy. Our network is the first to be simultaneously trained on three different datasets from the intelligent vehicles domain, i.e. Cityscapes, GTSDB and Mapillary Vistas, and is able to handle different semantic levelof-detail, class imbalances, and different annotation types, i.e. dense per-pixel and sparse bounding-box labels. We assess our hierarchical approach, by comparing against flat, nonhierarchical classifiers and we show improvements in mean pixel accuracy of 13.0% for Cityscapes classes and 2.4% for Vistas classes and 32.3% for GTSDB classes. Our implementation achieves inference rates of 17 fps at a resolution of 520 x 706 for 108 classes running on a GPU.

AB - We propose a convolutional network with hierarchical classifiers for per-pixel semantic segmentation, which is able to be trained on multiple, heterogeneous datasets and exploit their semantic hierarchy. Our network is the first to be simultaneously trained on three different datasets from the intelligent vehicles domain, i.e. Cityscapes, GTSDB and Mapillary Vistas, and is able to handle different semantic levelof-detail, class imbalances, and different annotation types, i.e. dense per-pixel and sparse bounding-box labels. We assess our hierarchical approach, by comparing against flat, nonhierarchical classifiers and we show improvements in mean pixel accuracy of 13.0% for Cityscapes classes and 2.4% for Vistas classes and 32.3% for GTSDB classes. Our implementation achieves inference rates of 17 fps at a resolution of 520 x 706 for 108 classes running on a GPU.

UR - http://www.scopus.com/inward/record.url?scp=85056747264&partnerID=8YFLogxK

U2 - 10.1109/IVS.2018.8500398

DO - 10.1109/IVS.2018.8500398

M3 - Conference contribution

AN - SCOPUS:85056747264

SP - 1045

EP - 1050

BT - 2018 IEEE Intelligent Vehicles Symposium, IV 2018

PB - Institute of Electrical and Electronics Engineers

ER -

Meletis P, Dubbelman G. Training of convolutional networks on multiple heterogeneous datasets for street scene semantic segmentation. In 2018 IEEE Intelligent Vehicles Symposium, IV 2018. Institute of Electrical and Electronics Engineers. 2018. blz. 1045-1050. 8500398 https://doi.org/10.1109/IVS.2018.8500398