EPFL / Cité des Sciences - Paris

Developed by Odoma for the Terra Data Exhibition in Paris, in partnership with the EPFL Digital Humanities Laboratory, this application allows users to navigate through over 30’000 artworks relying on deep neural networks.

Users can select one or more images to display all their related images, in an interactive exploration of the entire collection.

The interface was designed as an installation, therefore to be extremely robust yet very simple to understand and use.

The images are presented as a river, flowing slowly across the screen, giving visitors time to choose something of interest to interact with. The feed is updated and images similar to the one selected by users are displayed in a new flow.

This very simple operation demonstrates the power of the algorithm to identify visual themes across a wide and varied collection. For example, from a portrait painted in oil, the system will suggest similar contents over different supports, including sculpted busts, drawings or engravings. Therefore the ambitious goal has been not retrieving identical images but instead similar compositions, resulting in a creative and stimulating experience.

Added value:

• Search through artworks by visual similarity

• Find unexpected connections and themes

• Engage the public with a playful experience