Benchmarking Sampling-based Probabilistic Object Detectors.

Dimity Miller, Niko Sünderhauf, Haoyang Zhang, David Hall, Feras Dayoub

Abstract

This paper provides the first benchmark for sampling- based probabilistic object detectors. A probabilistic object detector expresses uncertainty for all detections that reliably indicates object localisation and classification performance. We compare performance for two sampling-based uncertainty techniques, namely Monte Carlo Dropout and Deep Ensembles, when implemented into one-stage and two-stage object detectors, Single Shot MultiBox Detector and Faster R-CNN. Our results show that Deep Ensembles outperform MC Dropout for both types of detectors. We also introduce a new merging strategy for sampling-based techniques and one-stage object detectors. We show this novel merging strategy has competitive performance with previously established strategies, while only having one free parameter.
BibTeX
@inproceedings{miller2019benchmarking,
  author = {Dimity Miller and Niko Sünderhauf and Haoyang Zhang and David Hall and Feras Dayoub},
  title = {{Benchmarking Sampling-based Probabilistic Object Detectors.}},
  booktitle = {Workshops @ CVPR},
  year = {2019},
}

Loading related content...

Figure