BayesOD: A Bayesian Approach for Uncertainty Estimation in Deep Object\n Detectors

Abstract

When incorporating deep neural networks into robotic systems, a major\nchallenge is the lack of uncertainty measures associated with their output\npredictions. Methods for uncertainty estimation in the output of deep object\ndetectors (DNNs) have been proposed in recent works, but have had limited\nsuccess due to 1) information loss at the detectors non-maximum suppression\n(NMS) stage, and 2) failure to take into account the multitask, many-to-one\nnature of anchor-based object detection. To that end, we introduce BayesOD, an\nuncertainty estimation approach that reformulates the standard object detector\ninference and Non-Maximum suppression components from a Bayesian perspective.\nExperiments performed on four common object detection datasets show that\nBayesOD provides uncertainty estimates that are better correlated with the\naccuracy of detections, manifesting as a significant reduction of\n9.77\\%-13.13\\% on the minimum Gaussian uncertainty error metric and a reduction\nof 1.63\\%-5.23\\% on the minimum Categorical uncertainty error metric. Code will\nbe released at {\\url{https://github.com/asharakeh/bayes-od-rc}}.\n