Bayesian SegNet: Model Uncertainty in Deep Convolutional Encoder-Decoder Architectures for Scene Understanding

TL;DR

Bayesian SegNet introduces a probabilistic deep learning framework for pixel-wise semantic segmentation that quantifies model uncertainty, improving accuracy especially on smaller datasets.

Contribution

It presents a practical Bayesian approach using dropout at test time to estimate uncertainty in segmentation models without extra parameters.

Findings

Uncertainty modeling improves segmentation accuracy by 2-3%.

Performance gains are more significant on smaller datasets.

The method is validated on SUN and CamVid datasets.

Abstract

We present a deep learning framework for probabilistic pixel-wise semantic segmentation, which we term Bayesian SegNet. Semantic segmentation is an important tool for visual scene understanding and a meaningful measure of uncertainty is essential for decision making. Our contribution is a practical system which is able to predict pixel-wise class labels with a measure of model uncertainty. We achieve this by Monte Carlo sampling with dropout at test time to generate a posterior distribution of pixel class labels. In addition, we show that modelling uncertainty improves segmentation performance by 2-3% across a number of state of the art architectures such as SegNet, FCN and Dilation Network, with no additional parametrisation. We also observe a significant improvement in performance for smaller datasets where modelling uncertainty is more effective. We benchmark Bayesian SegNet on the indoor SUN Scene Understanding and outdoor CamVid driving scenes datasets.