Breadcrumbs: Adversarial Class-Balanced Sampling for Long-tailed Recognition
Breadcrumbs: Adversarial Class-Balanced Sampling for Long-tailed Recognition
复制标题
DOI：
10.1007/978-3-031-20053-3_37
复制
发表时间：
2021-05
期刊：
影响因子：
--
通讯作者：
Bo Liu;Haoxiang Li;Hao Kang;G. Hua;N. Vasconcelos
中科院分区：
文献类型：
--
作者：
Bo Liu;Haoxiang Li;Hao Kang;G. Hua;N. Vasconcelos
长尾识别的问题，每类的例子的数量是高度不平衡的，被认为是。虽然使用类平衡采样的训练已被证明对这个问题有效，但众所周知，它会过度拟合少数几个类。据推测，这是由于重复采样的例子，可以解决的特征空间增强。提出了一种新的特征增强策略EMANATE，该策略基于训练过程中跨时期的特征回溯。结果表明，与类平衡抽样不同，这是一种对抗性增强策略。然后引入一个新的采样过程，面包屑，实现对抗类平衡采样，而无需额外的计算。在三个流行的长尾识别数据集上的实验表明，面包屑训练产生的分类器优于现有的问题解决方案。代码：https://github.com/BoLiu-SVCL/Breadcrumbs。
The problem of long-tailed recognition, where the number of examples per class is highly unbalanced, is considered. While training with class-balanced sampling has been shown effective for this problem, it is known to over-fit to few-shot classes. It is hypothesized that this is due to the repeated sampling of examples and can be addressed by feature space augmentation. A new feature augmentation strategy, EMANATE, based on back-tracking of features across epochs during training, is proposed. It is shown that, unlike class-balanced sampling, this is an adversarial augmentation strategy. A new sampling procedure, Breadcrumb, is then introduced to implement adversarial class-balanced sampling without extra computation. Experiments on three popular long-tailed recognition datasets show that Breadcrumb training produces classifiers that outperform existing solutions to the problem. Code: https://github.com/BoLiu-SVCL/Breadcrumbs.