Evaluating the Performance of Ensemble Methods and Voting Strategies for Dense 2D Pedestrian Detection in the Wild
Abstract
As vehicles experience a wide variety of driving settings in the wild, 2D pedestrian detection offers a substantial barrier to autonomous vehicle navigation systems. In this work, we demonstrate the effectiveness of a lightweight ensemble architecture for pedestrian detection in the wild, which combines detectors and data augmentation techniques to improve the performance of well-established detectors. The framework uses voting strategies to increase the explainability of object detection in navigation systems while also improving the precision of bounding box predictions on the dataset. The ensemble of the best model and augmentation technique achieved 41.41 % AP in detecting pedestrians in the wild using the consensus voting strategy on the WiderPerson dataset.
Cite
Text
Marathe et al. "Evaluating the Performance of Ensemble Methods and Voting Strategies for Dense 2D Pedestrian Detection in the Wild." IEEE/CVF International Conference on Computer Vision Workshops, 2021. doi:10.1109/ICCVW54120.2021.00398Markdown
[Marathe et al. "Evaluating the Performance of Ensemble Methods and Voting Strategies for Dense 2D Pedestrian Detection in the Wild." IEEE/CVF International Conference on Computer Vision Workshops, 2021.](https://mlanthology.org/iccvw/2021/marathe2021iccvw-evaluating/) doi:10.1109/ICCVW54120.2021.00398BibTeX
@inproceedings{marathe2021iccvw-evaluating,
title = {{Evaluating the Performance of Ensemble Methods and Voting Strategies for Dense 2D Pedestrian Detection in the Wild}},
author = {Marathe, Aboli and Walambe, Rahee and Kotecha, Ketan},
booktitle = {IEEE/CVF International Conference on Computer Vision Workshops},
year = {2021},
pages = {3568-3577},
doi = {10.1109/ICCVW54120.2021.00398},
url = {https://mlanthology.org/iccvw/2021/marathe2021iccvw-evaluating/}
}