Image Clustering

Image clustering aims to discover meaningful groups in unlabeled images, supporting tasks such as organizing photo collections, analyzing tissue images, and identifying species or objects. Common approaches learn image embeddings with convolutional networks or transformers, then use methods such as K-means, density-based clustering, spectral or subspace clustering, and graph construction. Newer methods also use language models and vision-language features to guide grouping, while hyperspectral and fine-grained settings incorporate spatial or domain-specific information. A central challenge is making clusters reliable across datasets and shifts while scaling computation and determining the number of groups; language guidance can help, but its benefits vary by dataset.