Abstract
Clustering artworks based on style can have many potential real-worldapplications like art recommendations, style-based search and retrieval, andthe study of artistic style evolution of an artist or in an artwork corpus. Weintroduce and deliberate over the notion of 'Style-based clustering of visualartworks'. We argue that clustering artworks based on style is largely anunaddressed problem. We explore and devise different neural featurerepresentations - from the style-classification, style-transfer to largelanguage vision models - that can be then used for style-based clustering. Ourobjective is to assess the relative effectiveness of these devised style-basedclustering approaches through qualitative and quantitative analysis by applyingthem to multiple artwork corpora and curated synthetically styled datasets.Besides providing a broad framework for style-based clustering and evaluation,our analysis provides some key novel insights on feature representations,architectures and implications for style-based clustering.