This website requires JavaScript.

Uncertainty-aware Panoptic Segmentation

Kshitij SirohiSajad MarviDaniel B\"uscherWolfram Burgard
Jun 2022
摘要
Reliable scene understanding is indispensable for modern autonomous systems.Current learning-based methods typically try to maximize their performancebased on segmentation metrics that only consider the quality of thesegmentation. However, for the safe operation of a system in the real world itis crucial to consider the uncertainty in the prediction as well. In this work,we introduce the novel task of uncertainty-aware panoptic segmentation, whichaims to predict per-pixel semantic and instance segmentations, together withper-pixel uncertainty estimates. We define two novel metrics to facilitate itsquantitative analysis, the uncertainty-aware Panoptic Quality (uPQ) and thepanoptic Expected Calibration Error (pECE). We further propose the noveltop-down Evidential Panoptic Segmentation Network (EvPSNet) to solve this task.Our architecture employs a simple yet effective probabilistic fusion modulethat leverages the predicted uncertainties. Additionally, we propose a newLov\'asz evidential loss function to optimize the IoU for the segmentationutilizing the probabilities provided by deep evidential learning. Furthermore,we provide several strong baselines combining state-of-the-art panopticsegmentation networks with sampling-free uncertainty estimation techniques.Extensive evaluations show that our EvPSNet achieves the new state-of-the-artfor the standard Panoptic Quality (PQ), as well as for our uncertainty-awarepanoptic metrics.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答