Simple and Robust Loss Design for Multi-Label Learning with Missing Labels

Abstract

Multi-label learning in the presence of missing labels (MLML) is a challenging problem. Existing methods mainly focus on the design of network structures or training schemes, which increase the complexity of implementation. This work seeks to fulfill the potential of loss function in MLML without increasing the procedure and complexity. Toward this end, we propose two simple yet effective methods via robust loss design based on an observation that a model can identify missing labels during training with a high precision. The first is a novel robust loss for negatives, namely the Hill loss, which re-weights negatives in the shape of a hill to alleviate the effect of false negatives. The second is a self-paced loss correction (SPLC) method, which uses a loss derived from the maximum likelihood criterion under an approximate distribution of missing labels. Comprehensive experiments on a vast range of multi-label image classification datasets demonstrate that our methods can remarkably boost the performance of MLML and achieve new state-of-the-art loss functions in MLML. Codes are available at https://github.com/xinyu1205/robust-loss-mlml.

Spatial Consistency Loss for Training Multi-Label Classifiers from Single-Label Annotations

“…We use the codebase shared by [9] to report the performance of ROLE with our setup. Comparison with [59] is in appendix B as it uses a different data split, which also includes partial labeling experiments where 40% or 75% of the positives are labeled instead of only a single positive.…”

“…The output of one serves as ground-truth for the other, with the intuition that both are more likely to converge to the same solution. Other approaches reweight samples based on their loss values [59,42]. Large Loss Matters [25] marks elements with large loss values as mislabeled and ignores or reweights those.…”

Knowledge Distillation from Single to Multi Labels: an Empirical Study

“…Besides, label correlation modeling [8,9,55,57] and the utilization of region features [15,36,50] are proved to be effective for multi-label classification. In light of the challenge of annotating all groundtruth labels for an image, multi-label learning in the presence of missing labels (MLML) has also attracted much research attention [11,20,56,61].…”

Spatial Consistency Loss for Training Multi-Label Classifiers from Single-Label Annotations

“…Ignoring the unannotated classes in the loss function can alleviate this issue [12], but this is inapplicable when the annotations only contain positives [8]. Asymmetric loss design can help handle missing labels beyond the BCE loss [54].…”

“…Prior works consider single-positive labels in the single-label setting [8,54], as a combination of single-label learning [37,11,20] and positive-unlabeled learning [10,1]. [8] propose to go beyond label smoothing [42,47] to deal with the label noise introduced by false negative labels their regularized online label estimation (ROLE) method estimates the missing labels in an online fashion, by jointly optimizing a label estimator and image classifier, the output of the former serving as ground truth of the latter.…”