SCAN: Learning to Classify Images Without Labels

Gansbeke, Wouter Van; Vandenhende, Simon; Georgoulis, Stamatios; Proesmans, Marc; Gool, Luc Van

doi:10.1007/978-3-030-58607-2_16

Cited by 310 publications

(343 citation statements)

References 34 publications

(68 reference statements)

Supporting

Mentioning

340

Contrasting

Order By: Relevance

“…Some methods combine several unsupervised steps to achieve firstly a good representation and then a clustering (e.g. [41]). In most cases, this unsupervised training is achieved by generating its own labels, and therefore the methods are called self-supervised.…”

Section: A Training Strategiesmentioning

confidence: 99%

“…A method that uses different algorithms, losses, or datasets during the training but only uses unsupervised data X u has one stage (e.g. [41]). A method which uses X u and X l during the complete training has one stage (e.g.…”

Section: A Training Strategiesmentioning

confidence: 99%

“…However, if we have a lot of classes or new classes should be detected like in few-or zeroshot learning [38], [94], [99], [100] unsupervised methods will still have a lower labeling cost and be of high importance. This means future research has to investigate how the semi-supervised ideas can be transferred to unsupervised methods as in [14], [41] and to settings with many, an unknown or rising amount of classes like in [39], [96].…”

Section: Trend: How Much Supervision Is Needed?mentioning

confidence: 99%

See 2 more Smart Citations

A Survey on Semi-, Self- and Unsupervised Learning for Image Classification

et al. 2021

View full text Add to dashboard Cite

While deep learning strategies achieve outstanding results in computer vision tasks, one issue remains: The current strategies rely heavily on a huge amount of labeled data. In many real-world problems, it is not feasible to create such an amount of labeled training data. Therefore, it is common to incorporate unlabeled data into the training process to reach equal results with fewer labels. Due to a lot of concurrent research, it is difficult to keep track of recent developments. In this survey, we provide an overview of often used ideas and methods in image classification with fewer labels. We compare 34 methods in detail based on their performance and their commonly used ideas rather than a fine-grained taxonomy. In our analysis, we identify three major trends that lead to future research opportunities. 1. State-of-the-art methods are scaleable to real-world applications in theory but issues like class imbalance, robustness, or fuzzy labels are not considered. 2. The degree of supervision which is needed to achieve comparable results to the usage of all labels is decreasing and therefore methods need to be extended to settings with a variable number of classes. 3. All methods share some common ideas but we identify clusters of methods that do not share many ideas. We show that combining ideas from different clusters can lead to better performance.

show abstract

Section: A Training Strategiesmentioning

confidence: 99%

Section: A Training Strategiesmentioning

confidence: 99%

Section: Trend: How Much Supervision Is Needed?mentioning

confidence: 99%

See 1 more Smart Citation

A Survey on Semi-, Self- and Unsupervised Learning for Image Classification

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Since our main contribution unravels the connection between Gaussian mixture models and autoencoders, models based on associative (Haeusser et al, 2018;Yang et al, 2019), spectral (Tian et al, 2014;Yang et al, 2019;Bianchi et al, 2020) or subspace (Zhang et al, 2019;Miklautz et al, 2020) clusterings are outside the scope of this paper. Furthermore, unlike (Yang et al, 2016;Chang et al, 2017;Kampffmeyer et al, 2019;Ghasedi Dizaji et al, 2017;Caron et al, 2018;Ji et al, 2019;Van Gansbeke et al, 2020), our model falls within the category of general purpose deep embedded clustering models which builds upon GMM (Xie et al, 2016;Yang et al, 2017;Fard et al, 2020). We mention that recent works related to specific topics such as perturbation robustness and non-redundant subspace clustering also utilize ideas that are to some degree related to the general concept of deep embedded clustering (Yang et al, 2020;Miklautz et al, 2020).…”

Section: Related Workmentioning

confidence: 99%

Joint optimization of an autoencoder for clustering and embedding

et al. 2021

View full text Add to dashboard Cite

Deep embedded clustering has become a dominating approach to unsupervised categorization of objects with deep neural networks. The optimization of the most popular methods alternates between the training of a deep autoencoder and a k-means clustering of the autoencoder’s embedding. The diachronic setting, however, prevents the former to benefit from valuable information acquired by the latter. In this paper, we present an alternative where the autoencoder and the clustering are learned simultaneously. This is achieved by providing novel theoretical insight, where we show that the objective function of a certain class of Gaussian mixture models (GMM’s) can naturally be rephrased as the loss function of a one-hidden layer autoencoder thus inheriting the built-in clustering capabilities of the GMM. That simple neural network, referred to as the clustering module, can be integrated into a deep autoencoder resulting in a deep clustering model able to jointly learn a clustering and an embedding. Experiments confirm the equivalence between the clustering module and Gaussian mixture models. Further evaluations affirm the empirical relevance of our deep architecture as it outperforms related baselines on several data sets.

show abstract

“…We show that the representations generated by MIX'EM result in high accuracy semantic clustering only by applying K-means to them. Existing works on unsupervised image classification using self-supervised representations [36,37,38] should benefit from adapting our proposed module, as it is an internal module that can be plugged-in without altering the output mechanism.…”

Section: Related Workmentioning

confidence: 99%

MIX’EM: Unsupervised Image Classification Using a Mixture of Embeddings

Varamesh

Tuytelaars

2021

Computer Vision – ACCV 2020

View full text Add to dashboard Cite

We present MIX'EM, a novel solution for unsupervised image classification. MIX'EM generates representations that by themselves are sufficient to drive a general-purpose clustering algorithm to deliver high-quality classification. This is accomplished by building a mixture of embeddings module into a contrastive visual representation learning framework in order to disentangle representations at the category level. It first generates a set of embedding and mixing coefficients from a given visual representation, and then combines them into a single embedding. We introduce three techniques to successfully train MIX'EM and avoid degenerate solutions; (i) diversify mixture components by maximizing entropy, (ii) minimize instance conditioned component entropy to enforce a clustered embedding space, and (iii) use an associative embedding loss to enforce semantic separability. By applying (i) and (ii), semantic categories emerge through the mixture coefficients, making it possible to apply (iii). Subsequently, we run K-means on the representations to acquire semantic classification. We conduct extensive experiments and analyses on STL10, CIFAR10, and CIFAR100-20 datasets, achieving state-of-theart classification accuracy of 78%, 82%, and 44%, respectively. To achieve robust and high accuracy, it is essential to use the mixture components to initialize K-means. Finally, we report competitive baselines (70% on STL10) obtained by applying K-means to the "normalized" representations learned using the contrastive loss.

show abstract

SCAN: Learning to Classify Images Without Labels

Cited by 310 publications

References 34 publications

A Survey on Semi-, Self- and Unsupervised Learning for Image Classification

A Survey on Semi-, Self- and Unsupervised Learning for Image Classification

Joint optimization of an autoencoder for clustering and embedding

MIX’EM: Unsupervised Image Classification Using a Mixture of Embeddings

Contact Info

Product

Resources

About