What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?

TLDR

Computer‑vision models face two main uncertainty types—aleatoric noise inherent in observations and epistemic uncertainty due to limited data—and while epistemic uncertainty has been hard to model, recent Bayesian deep‑learning tools now enable its estimation. This work investigates whether explicitly modeling epistemic versus aleatoric uncertainty improves performance in Bayesian deep‑learning vision tasks. The authors introduce a Bayesian framework that jointly estimates input‑dependent aleatoric uncertainty and epistemic uncertainty, and apply it to per‑pixel semantic segmentation and depth‑regression problems. Their uncertainty‑aware loss functions, viewed as learned attenuation, increase robustness to noisy data and yield new state‑of‑the‑art results on segmentation and depth‑regression benchmarks.

Abstract

There are two major types of uncertainty one can model. Aleatoric uncertainty captures noise inherent in the observations. On the other hand, epistemic uncertainty accounts for uncertainty in the model -- uncertainty which can be explained away given enough data. Traditionally it has been difficult to model epistemic uncertainty in computer vision, but with new Bayesian deep learning tools this is now possible. We study the benefits of modeling epistemic vs. aleatoric uncertainty in Bayesian deep learning models for vision tasks. For this we present a Bayesian deep learning framework combining input-dependent aleatoric uncertainty together with epistemic uncertainty. We study models under the framework with per-pixel semantic segmentation and depth regression tasks. Further, our explicit uncertainty formulation leads to new loss functions for these tasks, which can be interpreted as learned attenuation. This makes the loss more robust to noisy data, also giving new state-of-the-art results on segmentation and depth regression benchmarks.