Folli, an AI-based solution of the National Heritage Board, is designed to automatically describe and systematise visual material in museum collections. One important and time-consuming part of describing images is naming the objects they contain. Folli, on the other hand, can automatically find people, houses, furniture and other objects in photographs. This, in turn, improves search capabilities and allows more detailed analyses of the content of collections.

One part of the development of Folli is a demo application that visualises a collection of more than 250 000 photos based on visual similarity. Photos with similar content are placed closer together in the photo cloud, thus forming clusters based on visual themes. For example, in one part of the cloud you might find Soviet-era photos of a birthday party, in another part of the cloud you might find photos of a zeppelin whizzing over Tallinn, and so on.

You can find the demo application here: http://folli.stacc.cloud/demo
(the application may take a few minutes to start)

Folli uses a numeric vector shape to compare images. The images are processed using the InceptionV3 artificial neural network and the UMAP dimensionality reduction algorithm. The demo application's user interface uses PixPlot, a solution developed at Yale University, for the final visualization.

STACC OÜ is behind the technical implementation of the application. The work has been commissioned by the National Heritage Board of Estonia in cooperation with the National Library and the project has been funded by the European Regional Development Fund.