Benefit from large-scale training datasets, deep Convolutional Neural Networks(CNNs) have achieved impressive results in face recognition(FR). However, tremendous scale of datasets inevitably lead to noisy data, which obviously reduce the performance of the trained CNN models. Kicking out wrong labels from large-scale FR datasets is still very expensive, even some cleaning approaches are proposed. According to the analysis of the whole process of training CNN models supervised by angular margin based loss(AM-Loss) functions, we find that the $\theta$ distribution of training samples implicitly reflects their probability of being clean. Thus, we propose a novel training paradigm that employs the idea of weighting samples based on the above probability. Without any prior knowledge of noise, we can train high performance CNN models with large-scale FR datasets. Experiments demonstrate the effectiveness of our training paradigm. The codes are available at https://github.com/huangyangyu/NoiseFace.
翻译:从大规模培训数据集中受益,深革命神经网络(CNN)在面部识别方面取得了令人印象深刻的成果。然而,庞大的数据集不可避免地导致数据噪音,这显然降低了受过训练的CNN模型的性能。从大规模FR数据集中剔除错误标签仍然非常昂贵,甚至提出了一些清洁方法。根据对受以角差值为基础的损失功能(AM-Los)监督的CNN模型的整个培训过程的分析,我们发现,培训样本的美元分布暗含着其清洁的可能性。因此,我们提出了一个新的培训模式,利用基于上述概率的加权样品的概念。我们事先不了解噪音,就可以用大型FR数据集培训高性能CNN模型。实验表明我们培训模式的有效性。代码见https://github.com/huangyangyangyou/NoiseFace。