An Empirical Study into Clustering of Unseen Datasets with Self-Supervised Encoders

TL;DR

This study evaluates how well pretrained image models, especially self-supervised ones, can generalize to unseen datasets through clustering, revealing differences in feature representations and proposing silhouette score as a proxy for clustering quality.

Contribution

It provides a comprehensive benchmarking of pretrained encoders on unseen datasets, highlighting differences between supervised and self-supervised models and introducing silhouette score as a clustering performance proxy.

Findings

Supervised encoders perform better within training domain.

Self-supervised encoders excel on unseen datasets.

Silhouette score in UMAP space correlates with clustering performance.

Abstract

Can pretrained models generalize to new datasets without any retraining? We deploy pretrained image models on datasets they were not trained for, and investigate whether their embeddings form meaningful clusters. Our suite of benchmarking experiments use encoders pretrained solely on ImageNet-1k with either supervised or self-supervised training techniques, deployed on image datasets that were not seen during training, and clustered with conventional clustering algorithms. This evaluation provides new insights into the embeddings of self-supervised models, which prioritize different features to supervised models. Supervised encoders typically offer more utility than SSL encoders within the training domain, and vice-versa far outside of it, however, fine-tuned encoders demonstrate the opposite trend. Clustering provides a way to evaluate the utility of self-supervised learned representations orthogonal to existing methods such as kNN. Additionally, we find the silhouette score when measured in a UMAP-reduced space is highly correlated with clustering performance, and can therefore be used as a proxy for clustering performance on data with no ground truth labels. Our code implementation is available at \url{https://github.com/scottclowe/zs-ssl-clustering/}.