Deep Deterministic Uncertainty for Semantic Segmentation
This work addresses uncertainty estimation for semantic segmentation, which is important for safety-critical applications like autonomous driving, but it is incremental as it extends an existing method to a new domain.
The authors tackled uncertainty estimation in semantic segmentation by extending Deep Deterministic Uncertainty (DDU) to this task, enabling quantification of epistemic and aleatoric uncertainty in a single forward pass and reducing memory usage through location-independent application, with results showing improvements over MC Dropout and Deep Ensembles on Pascal VOC 2012 using DeepLab-v3+.
We extend Deep Deterministic Uncertainty (DDU), a method for uncertainty estimation using feature space densities, to semantic segmentation. DDU enables quantifying and disentangling epistemic and aleatoric uncertainty in a single forward pass through the model. We study the similarity of feature representations of pixels at different locations for the same class and conclude that it is feasible to apply DDU location independently, which leads to a significant reduction in memory consumption compared to pixel dependent DDU. Using the DeepLab-v3+ architecture on Pascal VOC 2012, we show that DDU improves upon MC Dropout and Deep Ensembles while being significantly faster to compute.