Contrastive Hierarchical Clustering

3 Mar 2023  ·  Michał Znaleźniak, Przemysław Rola, Patryk Kaszuba, Jacek Tabor, Marek Śmieja ·

Deep clustering has been dominated by flat models, which split a dataset into a predefined number of groups. Although recent methods achieve an extremely high similarity with the ground truth on popular benchmarks, the information contained in the flat partition is limited. In this paper, we introduce CoHiClust, a Contrastive Hierarchical Clustering model based on deep neural networks, which can be applied to typical image data. By employing a self-supervised learning approach, CoHiClust distills the base network into a binary tree without access to any labeled data. The hierarchical clustering structure can be used to analyze the relationship between clusters, as well as to measure the similarity between data points. Experiments demonstrate that CoHiClust generates a reasonable structure of clusters, which is consistent with our intuition and image semantics. Moreover, it obtains superior clustering accuracy on most of the image datasets compared to the state-of-the-art flat clustering models.

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Clustering CIFAR-10 CoHiClust Accuracy 0.839 # 18
NMI 0.779 # 14
Train set Train # 1
ARI 0.731 # 16
Backbone ResNet-50 # 1
Image Clustering CIFAR-100 CoHiClust Accuracy 0.437 # 17
NMI 0.467 # 14
ARI 0.299 # 15
Image Clustering Fashion-MNIST CoHiClust Accuracy 0.65 # 4
Image Clustering ImageNet-10 CoHiClust Accuracy 0.953 # 7
NMI 0.907 # 4
ARI 0.899 # 7
Backbone ResNet-50 # 1
Image Clustering Imagenet-dog-15 CoHiClust Accuracy 0.355 # 14
NMI 0.411 # 13
ARI 0.232 # 13
Backbone ResNet-50 # 1
Image Clustering MNIST CoHiClust Accuracy 0.99 # 2
Image Clustering STL-10 CoHiClust Accuracy 0.613 # 21
NMI 0.584 # 18
ARI 0.474 # 11
Backbone ResNet-50 # 1

Methods