Object Detection as Probabilistic Set Prediction

Accurate uncertainty estimates are essential for deploying deep object detectors in safety-critical systems. The development and evaluation of probabilistic object detectors have been hindered by shortcomings in existing performance measures, which tend to involve arbitrary thresholds or limit the detector's choice of distributions. In this work, we propose to view object detection as a set prediction task where detectors predict the distribution over the set of objects. Using the negative log-likelihood for random finite sets, we present a proper scoring rule for evaluating and training probabilistic object detectors. The proposed method can be applied to existing probabilistic detectors, is free from thresholds, and enables fair comparison between architectures. Three different types of detectors are evaluated on the COCO dataset. Our results indicate that the training of existing detectors is optimized toward non-probabilistic metrics. We hope to encourage the development of new object detectors that can accurately estimate their own uncertainty.

Automated summary

Accurate uncertainty estimation is crucial for deep object detectors in safety-critical systems. However, existing performance measures have limitations. This research proposes a novel approach that treats object detection as a set prediction task and presents a suitable scoring rule using negative log-likelihood for evaluating and training probabilistic object detectors. Evaluation on the COCO dataset reveals that current detectors' training is optimized towards non-probabilistic metrics. The study aims to inspire the development of new object detectors capable of accurately estimating their own uncertainty.