MetaDetect: Uncertainty Quantification and Prediction Quality Estimates for Object Detection

In object detection with deep neural networks, the box-wise objectness score tends to be overconfident, sometimes even indicating high confidence in presence of incorrect predictions. Hence, the reliability of the prediction and therefore reliable uncertainties are of highest interest. In this work, we present a post processing method that for any given neural network provides predictive uncertainty and quality estimates. These estimates are learned by a post processing model that receives as input a hand-crafted set of transparent metrics in form of a structured dataset. From this set we learn two tasks for predicted bounding boxes. We discriminate between true positives (IoU >= 0.5) and false positives (IoU < 0.5) which we term meta classification, and we predict IoU values directly which we term meta regression. The probabilities of the meta classification model aim to learn the probabilities of success and failure and therefore provide a modelled predictive uncertainty estimate. Moreover, meta regression gives rise to a quality estimate. In numerical experiments, we use the publicly available YOLOv3 network and the Faster-RCNN network and evaluate meta classification and regression performance on the Kitti, Pascal VOC and COCO datasets. We demonstrate that our metrics are indeed well correlated with the IoU. For meta classification we obtain classification accuracies of up to 99.61% and AUROCs of up to 99.84%. For meta regression we obtain an R-2 value of up to 91.78%. These results yield significant improvements compared to the network's objectness score and other baseline approaches. Therefore, we obtain more reliable uncertainty and quality estimates which is particularly interesting in the absence of ground truth.