Bayesian optimisation of existing object detection methods for new contexts
- Author
- Tim Willems (UGent) , Jan Aelterman (UGent) and David Van Hamme (UGent)
- Organization
- Project
- Abstract
- Pre-trained object detectors exhibit strong variations in performance when applied in different contexts, e.g., daytime vs. nighttime performance, up-close vs. long range detection. These variations limit the usefulness of pre-trained detectors for safety critical applications like autonomous driving, which require consistent decision making in every context. Retraining for all contexts is often impossible or prohibitively expensive due to the need for large amounts of labels in each context. Instead, we propose a probabilistic calibration layer which takes these context dependencies into account to translate the detection score produced by a pre-trained detector into a conditional probability of presence. As a proof of concept, we demonstrate that reinterpreting the confidence scores of three commonly used detectors based on the estimated distance to the supposed object yields an improvement in average precision of pedestrian detection of up to 3\% on the NuScenes dataset.
- Keywords
- Bayes, image processing, object detection, autonomous driving
Downloads
-
(...).pdf
- full text (Accepted manuscript)
- |
- UGent only
- |
- |
- 461.15 KB
Citation
Please use this url to cite or link to this publication: http://hdl.handle.net/1854/LU-01H8H9Q8C7E5PK7DDTEWQ3N7EX
- MLA
- Willems, Tim, et al. “Bayesian Optimisation of Existing Object Detection Methods for New Contexts.” 2023 IEEE SENSORS, IEEE, 2023, doi:10.1109/SENSORS56945.2023.10325317.
- APA
- Willems, T., Aelterman, J., & Van Hamme, D. (2023). Bayesian optimisation of existing object detection methods for new contexts. 2023 IEEE SENSORS. Presented at the IEEE SENSORS, Vienna, Austria. https://doi.org/10.1109/SENSORS56945.2023.10325317
- Chicago author-date
- Willems, Tim, Jan Aelterman, and David Van Hamme. 2023. “Bayesian Optimisation of Existing Object Detection Methods for New Contexts.” In 2023 IEEE SENSORS. IEEE. https://doi.org/10.1109/SENSORS56945.2023.10325317.
- Chicago author-date (all authors)
- Willems, Tim, Jan Aelterman, and David Van Hamme. 2023. “Bayesian Optimisation of Existing Object Detection Methods for New Contexts.” In 2023 IEEE SENSORS. IEEE. doi:10.1109/SENSORS56945.2023.10325317.
- Vancouver
- 1.Willems T, Aelterman J, Van Hamme D. Bayesian optimisation of existing object detection methods for new contexts. In: 2023 IEEE SENSORS. IEEE; 2023.
- IEEE
- [1]T. Willems, J. Aelterman, and D. Van Hamme, “Bayesian optimisation of existing object detection methods for new contexts,” in 2023 IEEE SENSORS, Vienna, Austria, 2023.
@inproceedings{01H8H9Q8C7E5PK7DDTEWQ3N7EX,
abstract = {{Pre-trained object detectors exhibit strong variations in performance when applied in different contexts, e.g., daytime vs. nighttime performance, up-close vs. long range detection. These variations limit the usefulness of pre-trained detectors for safety critical applications like autonomous driving, which require consistent decision making in every context. Retraining for all contexts is often impossible or prohibitively expensive due to the need for large amounts of labels in each context. Instead, we propose a probabilistic calibration layer which takes these context dependencies into account to translate the detection score produced by a pre-trained detector into a conditional probability of presence. As a proof of concept, we demonstrate that reinterpreting the confidence scores of three commonly used detectors based on the estimated distance to the supposed object yields an improvement in average precision of pedestrian detection of up to 3\% on the NuScenes dataset.}},
author = {{Willems, Tim and Aelterman, Jan and Van Hamme, David}},
booktitle = {{2023 IEEE SENSORS}},
isbn = {{9798350303872}},
issn = {{2168-9229}},
keywords = {{Bayes,image processing,object detection,autonomous driving}},
language = {{eng}},
location = {{Vienna, Austria}},
pages = {{4}},
publisher = {{IEEE}},
title = {{Bayesian optimisation of existing object detection methods for new contexts}},
url = {{http://doi.org/10.1109/SENSORS56945.2023.10325317}},
year = {{2023}},
}
- Altmetric
- View in Altmetric
- Web of Science
- Times cited: