Unsupervised Learning of Artistic Styles with Archetypal Style Analysis
Code: coming soon
We introduce an unsupervised learning approach to automatically discover, summarize, and manipulate artistic styles from large collections of paintings. Our method is based on archetypal analysis, which is an unsupervised learning technique akin to sparse coding with a geometric interpretation. When applied to deep image representations from a collection of artworks, it learns a dictionary of archetypal styles, which can be easily visualized. After training the model, the style of a new image, which is characterized by local statistics of deep visual features, is approximated by a sparse convex combination of archetypes. This enables us to interpret which archetypal styles are present in the input image, and in which proportion. Finally, our approach allows us to manipulate the coefficients of the latent archetypal decomposition, and achieve various special effects such as style enhancement, transfer, and interpolation between multiple archetypes.
Using deep archetypal style analysis, we can represent an artistic image (a) as a convex combination of archetypes. The archetypes can be visualized as synthesized textures (b), as a convex combination of artworks (c) or, when analyzing a specific image, as stylized versions of that image itself (d). Free recombination of the archetypal styles (e) then allows for novel stylizations of the input.
On these pages we present additional examples for our work. This includes comparisons of the \(\delta, \gamma\) style control to the control used by Li et. al. and results of our method as applied to the GanGogh collection, the collection of works by Vincent van Gogh and those by Claude Monet.
This work was supported by a grant from ANR (MACARON project under grant number ANR-14-CE23-0003-01), by the ERC grant number 714381 (SOLARIS project) and the ERC advanced grant Allegro.