SPICE: Semantic Pseudo-Labeling for Image Clustering.
SPICE: Semantic Pseudo-Labeling for Image Clustering.
复制标题
DOI：
10.1109/tip.2022.3221290
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
中科院分区：
文献类型：
--
作者：
样本间的相似性和类间的差异性是图像聚类的两个重要方面。然而，目前的深度聚类方法遭受不准确的特征相似性或语义差异的估计。在本文中，我们提出了一个基于语义伪标记的图像聚类（SPICE）框架，该框架将聚类网络划分为一个特征模型，用于测量实例级的相似性和聚类头，用于确定集群级的差异。我们设计了两个语义感知的伪标记算法，原型伪标记和可靠的伪标记，使准确和可靠的自我监督聚类。在不使用任何真实标签的情况下，我们分三个阶段优化聚类网络：1）通过对比学习来训练特征模型，以度量实例相似性; 2）使用原型伪标签算法来训练聚类头，以识别聚类语义; 3）使用可靠的伪标签算法联合训练特征模型和聚类头，以提高聚类性能。大量的实验结果表明，SPICE实现了显着的改进（~10%），现有的方法，并建立了新的国家的最先进的聚类结果在六个平衡的基准数据集方面的三个流行的指标。重要的是，SPICE显著减少了无监督分类和全监督分类之间的差距;例如，CIFAR-10的准确率仅为2%（91.8% vs 93.8%）。我们的代码可在https://github.com/niuchuangnn/SPICE上公开获取。
The similarity among samples and the discrepancy among clusters are two crucial aspects of image clustering. However, current deep clustering methods suffer from inaccurate estimation of either feature similarity or semantic discrepancy. In this paper, we present a Semantic Pseudo-labeling-based Image ClustEring (SPICE) framework, which divides the clustering network into a feature model for measuring the instance-level similarity and a clustering head for identifying the cluster-level discrepancy. We design two semantics-aware pseudo-labeling algorithms, prototype pseudo-labeling and reliable pseudo-labeling, which enable accurate and reliable self-supervision over clustering. Without using any ground-truth label, we optimize the clustering network in three stages: 1) train the feature model through contrastive learning to measure the instance similarity; 2) train the clustering head with the prototype pseudo-labeling algorithm to identify cluster semantics; and 3) jointly train the feature model and clustering head with the reliable pseudo-labeling algorithm to improve the clustering performance. Extensive experimental results demonstrate that SPICE achieves significant improvements (~10%) over existing methods and establishes the new state-of-the-art clustering results on six balanced benchmark datasets in terms of three popular metrics. Importantly, SPICE significantly reduces the gap between unsupervised and fully-supervised classification; e.g. there is only 2% (91.8% vs 93.8%) accuracy difference on CIFAR-10. Our code is made publicly available at https://github.com/niuchuangnn/SPICE.