Probabilistic Object Detection with an Ensemble of Experts

Abstract

Probabilistic object detection requires detectors to localise and classify objects in an image, while also providing accurate spatial and semantic uncertainty. In this work, we present an ‘Ensemble of Experts’ as a method to solve this challenging problem. This technique utilises a ranked ensembling association process and leverages the individual strengths of each expert detector to create a final set of detections with a meaningful spatial and semantic uncertainty. Our approach placed first place in the 3rd Probabilistic Object Detection Challenge with a PDQ of 22.848.

Cite

Text

Miller. "Probabilistic Object Detection with an Ensemble of Experts." European Conference on Computer Vision Workshops, 2020. doi:10.1007/978-3-030-65414-6_5

Markdown

[Miller. "Probabilistic Object Detection with an Ensemble of Experts." European Conference on Computer Vision Workshops, 2020.](https://mlanthology.org/eccvw/2020/miller2020eccvw-probabilistic/) doi:10.1007/978-3-030-65414-6_5

BibTeX

@inproceedings{miller2020eccvw-probabilistic,
  title     = {{Probabilistic Object Detection with an Ensemble of Experts}},
  author    = {Miller, Dimity},
  booktitle = {European Conference on Computer Vision Workshops},
  year      = {2020},
  pages     = {46-55},
  doi       = {10.1007/978-3-030-65414-6_5},
  url       = {https://mlanthology.org/eccvw/2020/miller2020eccvw-probabilistic/}
}