Dynamic Correlation Learning and Regularization for Multi-Label Confidence Calibration

Modern visual recognition models often display overconfidence due to their reliance on complex deep neural networks and one-hot target supervision, resulting in unreliable confidence scores that necessitate calibration. While current confidence calibration techniques primarily address single-label scenarios, there is a lack of focus on more practical and generalizable multi-label contexts. This paper introduces the Multi-Label Confidence Calibration (MLCC) task, aiming to provide well-calibrated confidence scores in multi-label scenarios. Unlike single-label images, multi-label images contain multiple objects, leading to semantic confusion and further unreliability in confidence scores. Existing single-label calibration methods, based on label smoothing, fail to account for category correlations, which are crucial for addressing semantic confusion, thereby yielding sub-optimal performance. To overcome these limitations, we propose the Dynamic Correlation Learning and Regularization (DCLR) algorithm, which leverages multi-grained semantic correlations to better model semantic confusion for adaptive regularization. DCLR learns dynamic instance-level and prototype-level similarities specific to each category, using these to measure semantic correlations across different categories. With this understanding, we construct adaptive label vectors that assign higher values to categories with strong correlations, thereby facilitating more effective regularization. We establish an evaluation benchmark, re-implementing several advanced confidence calibration algorithms and applying them to leading multi-label recognition (MLR) models for fair comparison. Through extensive experiments, we demonstrate the superior performance of DCLR over existing methods in providing reliable confidence scores in multi-label scenarios.

I. Introduction

Modern visual recognition models, built on complex deep neural networks [1], [2], often suffer from overfitting to training data, inevitably leading to overly confident and unreliable predicted score dilemma [3], [4], [5]. This dilemma severely prevents their applications to high-risk scenarios, such as self-driving [6], [7] and medical diagnosis [8], [9]. To deal with this issue, numerous works [3], [10], [11] are intensively proposed for confidence calibration that can provide more accurate and reliable predicted confidence scores to indicate an accurate probability of correctness. Despite achieving impressive progress, these efforts predominantly concentrate on single-label settings, where each image is associated with a single category. However, these works can hardly be applied to multi-label scenarios, which are more reflective of real-world scenarios where images often contain objects from multiple categories [12], [13], [14]. Our work targets the multi-label confidence calibration (MLCC) task, seeking to extend and enhance calibration techniques for these more complex and practical scenarios.