Publication detail

Pedestrian Detector Domain Shift Robustness Evaluation, and Domain Shift Error Mitigation Proposal

ZEMČÍK, T.

Original Title

Pedestrian Detector Domain Shift Robustness Evaluation, and Domain Shift Error Mitigation Proposal

Type

conference paper

Language

English

Original Abstract

This paper evaluates daytime to nighttime traffic image domain shift on Faster R-CNN and SSD based pedestrian and cyclist detectors. Daytime image trained detectors are applied on a newly compiled nighttime image dataset and their performance is evaluated against detectors trained on both daytime and nighttime images. Faster R-CNN based detectors proved relatively robust, but still clearly inferior to the models trained on nighttime images, the SSD based model proved noncompetitive. Approaches to the domain shift deterioration mitigation were proposed and future work outlined.

Keywords

Object detection, Pedestrian detection, Cyclist detection, ADAS, AV, Faster R-CNN, SSD, Domain shift, Domain adaptation, Data augmentation

Authors

ZEMČÍK, T.

Released

27. 4. 2021

Publisher

Brno University of Technology, Faculty of Electrical Engineering and Communication

Location

Brno

ISBN

978-80-214-5943-4

Book

Proceedings II of the 27th student EEICT selected papers

Edition

1

Pages from

181

Pages to

187

Pages count

7

URL

https://www.eeict.cz/eeict_download/archiv/sborniky/EEICT_2021_sbornik_2_v3_DOI.pdf

BibTex

@inproceedings{BUT164116,
  author="Tomáš {Zemčík}",
  title="Pedestrian Detector Domain Shift Robustness Evaluation, and Domain Shift Error Mitigation Proposal",
  booktitle="Proceedings II of the 27th student EEICT selected papers",
  year="2021",
  series="1",
  pages="181--187",
  publisher="Brno University of Technology, Faculty of Electrical Engineering and Communication",
  address="Brno",
  doi="10.13164/eeict.2021.181",
  isbn="978-80-214-5943-4",
  url="https://www.eeict.cz/eeict_download/archiv/sborniky/EEICT_2021_sbornik_2_v3_DOI.pdf"
}