|Cordelia Schmid||Inria / Google|
The authors introduce an unsupervised learning approach to automatically dis-cover, summarize, and manipulate artistic styles from large collections of paintings.
In this paper, we introduce an unsupervised learning approach to automatically dis-cover, summarize, and manipulate artistic styles from large collections of paintings.Our method is based on archetypal analysis, which is an unsupervised learningtechnique akin to sparse coding with a geometric interpretation. When appliedto deep image representations from a data collection, it learns a dictionary ofarchetypal styles, which can be easily visualized. After training the model, the styleof a new image, which is characterized by local statistics of deep visual features,is approximated by a sparse convex combination of archetypes. This allows usto interpret which archetypal styles are present in the input image, and in whichproportion. Finally, our approach allows us to manipulate the coefficients of thelatent archetypal decomposition, and achieve various special effects such as styleenhancement, transfer, and interpolation between multiple archetypes.