Abstract:Unsupervised image clustering (UIC) is regularly employed to group images without manual annotation. One significant problem that occurs in the UIC context is that the visual-feature similarity across different semantic classes tends to introduce instance-dependent errors to clustering. The most successful recent approaches aimed at resolving this problem have focused on semisupervised reclustering, which utilizes reliable samples selected from the existing clusters. Despite this, virtually no previous work has considered the spatial consistency of the instance- and class-level representations which is crucial for error disambiguation. This makes it difficult to assess whether the selected reliable sample is reasonable. Accordingly, we propose a spatial consistency-based clustering (SCC) method to retain the alignment of representations learned from the instance and class levels in order to effectively select reliable samples from pre-existing clusters with errors thus leading to better clustering performance. More specifically, we first learn instance- and class-level representations by encouraging the semantic invariants of different instance augmentations and enforcing class alignment across semantically similar instances, respectively. We then assign instances with similar class-level representations to the same cluster to obtain the preliminary clusters. Subsequently we assess sample reliability by utilizing the spatial consistency constraint, which diffuses the class-level representations within the instance-level representation space. Finally, we employ semisupervised baselines combined with refinement techniques to perform reclustering based on the selected reliable samples. Extensive experimental results demonstrate that SCC can effectively obtain credible samples and outperform current SOTA clustering methods on the CIFAR-10 and CIFAR-100-20 benchmarks. The relevant code is available at https://github.com/RyanZhaoIc/SCC.git. (C)2022 Elsevier B.V. All rights reserved.

Subspace Clustering Via Independent Subspace Analysis Network

Subspace Clustering Via Stacked Independent Subspace Analysis Networks With Sparse Prior Information

Learning a Self-Expressive Network for Subspace Clustering

Subspace Clustering Via Integrating Sparse Representation and Adaptive Graph Learning.

Structure-Aware Subspace Clustering

Multiple Independent Subspace Clusterings.

Deep Subspace Image Clustering Network with Self-Expression and Self-Supervision

Subspace Clustering for Vector Clusters

Exploiting Unsupervised and Supervised Constraints for Subspace Clustering

SSCNet: learning-based subspace clustering

Self-Paced Subspace Clustering

Attributed Subspace Clustering.

Subspace Clustering with Sparsity and Grouping Effect

Nonlinear Subspace Clustering Via Adaptive Graph Regularized Autoencoder.

Sparse Subspace Clustering with Jointly Learning Representation and Affinity Matrix.

Improving unsupervised image clustering with spatial consistency

Inductive Latent Space Sparse and Low-rank Subspace Clustering Algorithm

Self-Supervised Deep Multi-View Subspace Clustering.

Iterative Deep Subspace Clustering.

Inductive Sparse Subspace Clustering

Deep Structure and Attention Aware Subspace Clustering