CNN features are also great at unsupervised classification

TL;DR

This paper demonstrates that features extracted from pretrained deep CNNs are highly effective for unsupervised image clustering, outperforming many existing methods and applicable to robotic object sorting.

Contribution

It shows that simple transfer learning with pretrained CNN features combined with classic clustering surpasses state-of-the-art unsupervised algorithms.

Findings

Pretrained CNN features improve clustering accuracy.

The approach outperforms existing unsupervised methods.

Effective in robotic object sorting applications.

Abstract

This paper aims at providing insight on the transferability of deep CNN features to unsupervised problems. We study the impact of different pretrained CNN feature extractors on the problem of image set clustering for object classification as well as fine-grained classification. We propose a rather straightforward pipeline combining deep-feature extraction using a CNN pretrained on ImageNet and a classic clustering algorithm to classify sets of images. This approach is compared to state-of-the-art algorithms in image-clustering and provides better results. These results strengthen the belief that supervised training of deep CNN on large datasets, with a large variability of classes, extracts better features than most carefully designed engineering approaches, even for unsupervised tasks. We also validate our approach on a robotic application, consisting in sorting and storing objects smartly based on clustering.