Data selection for training semantic segmentation CNNs with cross-dataset weak supervision

Panagiotis Meletis, R.R.F.M. Romijnders, Gijs Dubbelman

Onderzoeksoutput: Hoofdstuk in Boek/Rapport/CongresprocedureConferentiebijdrageAcademicpeer review

Uittreksel

Training convolutional networks for semantic segmentation with strong (per-pixel) and weak (per-bounding-box) supervision requires a large amount of weakly labeled data. We propose two methods for selecting the most relevant data with weak supervision. The first method is designed for finding visually similar images without the need of labels and is based on modeling image representations with a Gaussian Mixture Model (GMM). As a byproduct of GMM modeling, we present useful insights on characterizing the data generating distribution. The second method aims at finding images with high object diversity and requires only the bounding box labels. Both methods are developed in the context of automated driving and experimentation is conducted on Cityscapes and Open Images datasets. We demonstrate performance gains by reducing the amount of employed weakly labeled images up to 100 times for Open Images and up to 20 times for Cityscapes.
Originele taal-2Engels
Titel2019 IEEE Intelligent Transportation Systems Conference (ITSC)
Plaats van productiePiscataway
UitgeverijInstitute of Electrical and Electronics Engineers
Pagina's3682-3688
Aantal pagina's7
ISBN van elektronische versie978-1-5386-7024-8
DOI's
StatusGepubliceerd - 2019
EvenementIEEE ITSC 2019 - Auckland, Nieuw-Zeeland
Duur: 27 okt 201930 okt 2019

Congres

CongresIEEE ITSC 2019
LandNieuw-Zeeland
StadAuckland
Periode27/10/1930/10/19

Vingerafdruk

Semantics
Labels
Byproducts
Pixels

Citeer dit

Meletis, P., Romijnders, R. R. F. M., & Dubbelman, G. (2019). Data selection for training semantic segmentation CNNs with cross-dataset weak supervision. In 2019 IEEE Intelligent Transportation Systems Conference (ITSC) (blz. 3682-3688). Piscataway: Institute of Electrical and Electronics Engineers. https://doi.org/10.1109/ITSC.2019.8917069
Meletis, Panagiotis ; Romijnders, R.R.F.M. ; Dubbelman, Gijs. / Data selection for training semantic segmentation CNNs with cross-dataset weak supervision. 2019 IEEE Intelligent Transportation Systems Conference (ITSC). Piscataway : Institute of Electrical and Electronics Engineers, 2019. blz. 3682-3688
@inproceedings{de4c842b0e8d436f8020239652868031,
title = "Data selection for training semantic segmentation CNNs with cross-dataset weak supervision",
abstract = "Training convolutional networks for semantic segmentation with strong (per-pixel) and weak (per-bounding-box) supervision requires a large amount of weakly labeled data. We propose two methods for selecting the most relevant data with weak supervision. The first method is designed for finding visually similar images without the need of labels and is based on modeling image representations with a Gaussian Mixture Model (GMM). As a byproduct of GMM modeling, we present useful insights on characterizing the data generating distribution. The second method aims at finding images with high object diversity and requires only the bounding box labels. Both methods are developed in the context of automated driving and experimentation is conducted on Cityscapes and Open Images datasets. We demonstrate performance gains by reducing the amount of employed weakly labeled images up to 100 times for Open Images and up to 20 times for Cityscapes.",
author = "Panagiotis Meletis and R.R.F.M. Romijnders and Gijs Dubbelman",
year = "2019",
doi = "10.1109/ITSC.2019.8917069",
language = "English",
pages = "3682--3688",
booktitle = "2019 IEEE Intelligent Transportation Systems Conference (ITSC)",
publisher = "Institute of Electrical and Electronics Engineers",
address = "United States",

}

Meletis, P, Romijnders, RRFM & Dubbelman, G 2019, Data selection for training semantic segmentation CNNs with cross-dataset weak supervision. in 2019 IEEE Intelligent Transportation Systems Conference (ITSC). Institute of Electrical and Electronics Engineers, Piscataway, blz. 3682-3688, IEEE ITSC 2019, Auckland, Nieuw-Zeeland, 27/10/19. https://doi.org/10.1109/ITSC.2019.8917069

Data selection for training semantic segmentation CNNs with cross-dataset weak supervision. / Meletis, Panagiotis; Romijnders, R.R.F.M.; Dubbelman, Gijs.

2019 IEEE Intelligent Transportation Systems Conference (ITSC). Piscataway : Institute of Electrical and Electronics Engineers, 2019. blz. 3682-3688.

Onderzoeksoutput: Hoofdstuk in Boek/Rapport/CongresprocedureConferentiebijdrageAcademicpeer review

TY - GEN

T1 - Data selection for training semantic segmentation CNNs with cross-dataset weak supervision

AU - Meletis, Panagiotis

AU - Romijnders, R.R.F.M.

AU - Dubbelman, Gijs

PY - 2019

Y1 - 2019

N2 - Training convolutional networks for semantic segmentation with strong (per-pixel) and weak (per-bounding-box) supervision requires a large amount of weakly labeled data. We propose two methods for selecting the most relevant data with weak supervision. The first method is designed for finding visually similar images without the need of labels and is based on modeling image representations with a Gaussian Mixture Model (GMM). As a byproduct of GMM modeling, we present useful insights on characterizing the data generating distribution. The second method aims at finding images with high object diversity and requires only the bounding box labels. Both methods are developed in the context of automated driving and experimentation is conducted on Cityscapes and Open Images datasets. We demonstrate performance gains by reducing the amount of employed weakly labeled images up to 100 times for Open Images and up to 20 times for Cityscapes.

AB - Training convolutional networks for semantic segmentation with strong (per-pixel) and weak (per-bounding-box) supervision requires a large amount of weakly labeled data. We propose two methods for selecting the most relevant data with weak supervision. The first method is designed for finding visually similar images without the need of labels and is based on modeling image representations with a Gaussian Mixture Model (GMM). As a byproduct of GMM modeling, we present useful insights on characterizing the data generating distribution. The second method aims at finding images with high object diversity and requires only the bounding box labels. Both methods are developed in the context of automated driving and experimentation is conducted on Cityscapes and Open Images datasets. We demonstrate performance gains by reducing the amount of employed weakly labeled images up to 100 times for Open Images and up to 20 times for Cityscapes.

U2 - 10.1109/ITSC.2019.8917069

DO - 10.1109/ITSC.2019.8917069

M3 - Conference contribution

SP - 3682

EP - 3688

BT - 2019 IEEE Intelligent Transportation Systems Conference (ITSC)

PB - Institute of Electrical and Electronics Engineers

CY - Piscataway

ER -

Meletis P, Romijnders RRFM, Dubbelman G. Data selection for training semantic segmentation CNNs with cross-dataset weak supervision. In 2019 IEEE Intelligent Transportation Systems Conference (ITSC). Piscataway: Institute of Electrical and Electronics Engineers. 2019. blz. 3682-3688 https://doi.org/10.1109/ITSC.2019.8917069