Get our free extension to see links to code for papers anywhere online!

Chrome logo Add to Chrome

Firefox logo Add to Firefox

Can We Trust You? On Calibration of a Probabilistic Object Detector for Autonomous Driving

Sep 26, 2019
Di Feng, Lars Rosenbaum, Claudius Glaeser, Fabian Timm, Klaus Dietmayer

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.

* To appear in IROS 2019 

Share this with someone who'll enjoy it:

   Access Paper Source

Share this with someone who'll enjoy it: