Uncertainty Estimation for 3D Dense Prediction via Cross-Point Embeddings

TL;DR

This paper introduces CUE and CUE+ methods for estimating uncertainty in 3D dense prediction tasks, leveraging cross-point embeddings to improve calibration without affecting accuracy.

Contribution

The paper proposes novel uncertainty estimation techniques, CUE and CUE+, that utilize cross-point embeddings and metric learning for 3D point cloud predictions.

Findings

CUE and CUE+ effectively calibrate uncertainty in 3D geometric feature learning.

CUE+ reduces the Expected Calibration Error in semantic segmentation by 16.5%.

Uncertainty estimation is achieved without sacrificing predictive performance.

Abstract

Dense prediction tasks are common for 3D point clouds, but the uncertainties inherent in massive points and their embeddings have long been ignored. In this work, we present CUE, a novel uncertainty estimation method for dense prediction tasks in 3D point clouds. Inspired by metric learning, the key idea of CUE is to explore cross-point embeddings upon a conventional 3D dense prediction pipeline. Specifically, CUE involves building a probabilistic embedding model and then enforcing metric alignments of massive points in the embedding space. We also propose CUE+, which enhances CUE by explicitly modeling crosspoint dependencies in the covariance matrix. We demonstrate that both CUE and CUE+ are generic and effective for uncertainty estimation in 3D point clouds with two different tasks: (1) in 3D geometric feature learning we for the first time obtain wellcalibrated uncertainty, and (2) in semantic segmentation we reduce uncertainty's Expected Calibration Error of the state-of-the-arts by 16.5%. All uncertainties are estimated without compromising predictive performance.