Prediction Quality Meta Regression and Error Meta Classification for Segmented Lidar Point Clouds

Abstract

We present a post-processing tool for semantic segmentation of Lidar point clouds, called LidarMetaSeg, which estimates the prediction quality segmentwise and classifies prediction errors. For this purpose, we compute dispersion measures based on network probability outputs as well as feature measures based on point cloud input features and aggregate them on segment level. These aggregated measures are used to train a meta classification model to predict whether a predicted segment is an error, i.e., it is a false positive or not. We also train a meta regression model to predict the segmentwise prediction quality in terms of intersection over union. Both models can then be applied to semantic segmentation inferences without knowing the ground truth. In our experiments we use different Lidar segmentation models and datasets and analyze the power of our method. We show that our results outperform other standard approaches based on single uncertainty measures like entropy. Furthermore, we present an in-depth evaluation of our method on predicted classes as well as on predicted categories.