Neighbor-Aware Calibration of Segmentation Networks with Penalty-Based Constraints

TL;DR

This paper introduces NACL, a neighbor-aware calibration method for segmentation networks that explicitly controls local structure constraints, improving confidence calibration without sacrificing discriminative ability.

Contribution

The paper presents NACL, a novel calibration approach based on explicit equality constraints on logits, addressing limitations of prior methods like SVLS by balancing constraint enforcement and primary objectives.

Findings

NACL outperforms existing calibration methods on various benchmarks.

The approach maintains discriminative power while improving confidence calibration.

NACL is model-agnostic and applicable to different segmentation architectures.

Abstract

Ensuring reliable confidence scores from deep neural networks is of paramount significance in critical decision-making systems, particularly in real-world domains such as healthcare. Recent literature on calibrating deep segmentation networks has resulted in substantial progress. Nevertheless, these approaches are strongly inspired by the advancements in classification tasks, and thus their uncertainty is usually modeled by leveraging the information of individual pixels, disregarding the local structure of the object of interest. Indeed, only the recent Spatially Varying Label Smoothing (SVLS) approach considers pixel spatial relationships across classes, by softening the pixel label assignments with a discrete spatial Gaussian kernel. In this work, we first present a constrained optimization perspective of SVLS and demonstrate that it enforces an implicit constraint on soft class proportions of surrounding pixels. Furthermore, our analysis shows that SVLS lacks a mechanism to balance the contribution of the constraint with the primary objective, potentially hindering the optimization process. Based on these observations, we propose NACL (Neighbor Aware CaLibration), a principled and simple solution based on equality constraints on the logit values, which enables to control explicitly both the enforced constraint and the weight of the penalty, offering more flexibility. Comprehensive experiments on a wide variety of well-known segmentation benchmarks demonstrate the superior calibration performance of the proposed approach, without affecting its discriminative power. Furthermore, ablation studies empirically show the model agnostic nature of our approach, which can be used to train a wide span of deep segmentation networks.