On the Pitfalls of Entropy-Based Uncertainty for Multi-class Semi-supervised Segmentation

Martin Van Waerebeke, Gregory A. Lodygensky, Jose Dolz

Abstract: Estimating the prediction uncertainty of a deep segmentation network is very useful in multiple learning scenarios. For example, in the semi-supervised learning paradigm, the vast majority of recent methods rely on pseudo-label generation to leverage unlabeled data, whose training is guided by uncertainty estimates. While the commonly-used entropy-based uncertainty has shown to work well in a binary scenario, we demonstrate in this work that this common strategy leads to suboptimal results in a multi-class context, a more realistic and challenging setting. We argue, indeed, that these approaches underperform due to the erroneous uncertainty approximations in the presence of inter-class overlap. Furthermore, we propose an alternative solution to compute the uncertainty in a multi-class setting, based on divergence distances and which account for inter-class overlap. We evaluate the proposed solution on a challenging multi-class segmentation dataset and in two well-known uncertainty-based segmentation methods. The reported results demonstrate that by simply replacing the mechanism used to compute the uncertainty, our proposed solution brings consistent improvements .