Abstract:Semantic segmentation (SS) aims to classify each pixel into one of the pre-defined classes. This task plays an important role in self-driving cars and autonomous drones. In SS, many works have shown that most misclassified pixels are commonly near object boundaries with high uncertainties. However, existing SS loss functions are not tailored to handle these uncertain pixels during training, as these pixels are usually treated equally as confidently classified pixels and cannot be embedded with arbitrary low distortion in Euclidean space, thereby degenerating the performance of SS. To overcome this problem, this paper designs a Hyperbolic Uncertainty Loss (HyperUL), which dynamically highlights the misclassified and high-uncertainty pixels in Hyperbolic space during training via the hyperbolic distances. The proposed HyperUL is model agnostic and can be easily applied to various neural architectures. After employing HyperUL to three recent SS models, the experimental results on Cityscapes, UAVid, and ACDC datasets reveal that the segmentation performance of existing SS models can be consistently improved. Additionally, reliable measurement of model uncertainty plays a key role in real-world applications such as autonomous controls of vehicles and drones. To meet this requirement, we propose the Hyperbolic Uncertainty Estimation method, which is easily implemented by only post-processing the generated Hyperbolic embeddings. By this approach, we can calculate the uncertainty values almost for free. Quantitative and qualitative results on Cityscapes, UAVid, and ACDC datasets verify that our proposed uncertainty estimation method usually outputs more meaningful results compared with popular MC-dropout and ensembling methods.

Embodied Uncertainty-Aware Object Segmentation

Uncertainty Visualization for Mobile and Wearable Devices Based Activity Recognition Systems

Perception Matters: Enhancing Embodied AI with Uncertainty-Aware Semantic Segmentation

Uncertainty estimation in Deep Learning for Panoptic segmentation

Hyperbolic Uncertainty Aware Semantic Segmentation

Unsupervised Online Video Object Segmentation with Motion Property Understanding

An Uncertainty Estimation Framework for Probabilistic Object Detection

Mitigating Distributional Shift in Semantic Segmentation via Uncertainty Estimation From Unlabeled Data

Uncertainty Estimation and Out-of-Distribution Detection for LiDAR Scene Semantic Segmentation

Mitigating Distributional Shift in Semantic Segmentation via Uncertainty Estimation from Unlabelled Data

Robot Unknown Objects Instance Segmentation Based on Collaborative Weight Assignment RGB–Depth Fusion Strategy

Anatomically-aware uncertainty for semi-supervised image segmentation

High-quality Unknown Object Instance Segmentation via Quadruple Boundary Error Refinement

Uncertainty in Real-Time Semantic Segmentation on Embedded Systems

RISeg: Robot Interactive Object Segmentation via Body Frame-Invariant Features

Learning 3D Human Pose and Shape Estimation Using Uncertainty-Aware Body Part Segmentation

Segmenting Known Objects and Unseen Unknowns without Prior Knowledge

Uncertainty Aware Proposal Segmentation for Unknown Object Detection

An Intuitionistic Fuzzy Set Driven Stochastic Active Contour Model with Uncertainty Analysis

Elongated Physiological Structure Segmentation via Spatial and Scale Uncertainty-aware Network

Instance-wise Uncertainty for Class Imbalance in Semantic Segmentation