Boosting Single Positive Multi-label Classification with Generalized Robust Loss
Yanxi Chen, Chunxiao Li, Xinyang Dai, Jinhuan Li, Weiyu Sun, Yiming Wang, Renyuan Zhang, Tinghe Zhang, Bo Wang
TL;DR
This paper tackles SPML where each sample has only one verified positive label, addressing the resultant false negatives and label imbalance. It introduces Generalized Robust Loss (GR Loss), a framework built on expected risk minimization that combines soft pseudo-labeling via a calibrated $\hat{k}(p;\beta)$, instance/class-aware reweighting $v(p;\alpha)$, and three robust surrogate losses $\mathcal{L}_1, \mathcal{L}_2, \mathcal{L}_3$ to form a unified loss. The method unifies and extends existing SPML/MLML losses by mapping them into the GR framework, and it demonstrates state-of-the-art performance on VOC, COCO, and NUS benchmarks, with competitive results on CUB due to label-correlation effects. The work provides gradient analyses and ablations that highlight the importance of pseudo-label calibration and robust loss design for handling missing labels and intra-/inter-class imbalance, offering practical gains for SPML applications while noting limitations and future directions such as dynamic loss parameters and explicit label correlations.
Abstract
Multi-label learning (MLL) requires comprehensive multi-semantic annotations that is hard to fully obtain, thus often resulting in missing labels scenarios. In this paper, we investigate Single Positive Multi-label Learning (SPML), where each image is associated with merely one positive label. Existing SPML methods only focus on designing losses using mechanisms such as hard pseudo-labeling and robust losses, mostly leading to unacceptable false negatives. To address this issue, we first propose a generalized loss framework based on expected risk minimization to provide soft pseudo labels, and point out that the former losses can be seamlessly converted into our framework. In particular, we design a novel robust loss based on our framework, which enjoys flexible coordination between false positives and false negatives, and can additionally deal with the imbalance between positive and negative samples. Extensive experiments show that our approach can significantly improve SPML performance and outperform the vast majority of state-of-the-art methods on all the four benchmarks.