作者
Di Feng, Lars Rosenbaum, Claudius Glaeser, Fabian Timm, Klaus Dietmayer
发表日期
2019/9/26
期刊
arXiv preprint arXiv:1909.12358
简介
Reliable uncertainty estimation is crucial for perception systems in safe autonomous driving. Recently, many methods have been proposed to model uncertainties in deep learning based object detectors. However, the estimated probabilities are often uncalibrated, which may lead to severe problems in safety critical scenarios. In this work, we identify such uncertainty miscalibration problems in a probabilistic LiDAR 3D object detection network, and propose three practical methods to significantly reduce errors in uncertainty calibration. Extensive experiments on several datasets show that our methods produce well-calibrated uncertainties, and generalize well between different datasets.
学术搜索中的文章
D Feng, L Rosenbaum, C Glaeser, F Timm… - arXiv preprint arXiv:1909.12358, 2019