An AI Identifies an Address from an Interior Photo

Graylark introduces Raven, a beta tool designed for investigators that suggests potential addresses from a simple interior photo, even without metadata or an exterior view. The system relies on available real estate images and displays a similarity percentage.

Announced Beta Functionality and Usage Conditions

The Find Property feature of Raven is currently in beta. It allows users to associate a room or facade with its address, but only in supported cities. After analysis, the tool proposes potential addresses along with a similarity rate expressed as a percentage, based on available property images. According to Graylark's description, location identification is possible even if the interior photo shows no view to the outside and contains no metadata.

A Tool Developed for Investigation, Cited by Malwarebytes

Raven was developed by Graylark, an American company that positions itself as an artificial intelligence laboratory creating AI and data-driven solutions for law enforcement and investigative teams. The software was recently mentioned by Malwarebytes, an antivirus publisher. Graylark describes Raven as a means to associate an address with an interior photo by cross-referencing clues with available real estate images.

Exploited Clues and Multimodal Approach

For analysis, Raven examines details such as wallpaper or tile patterns, furniture, and other visual elements. Its FAQ specifies that the tool compares the layout of rooms, finishes, fixtures, furniture, and other permanent visual details with available real estate images to identify candidate addresses. Graylark attributes these capabilities to the use of multimodal models and the processing of a large amount of data and files. Additionally, an AI related to ChatGPT demonstrated last year that an exterior photo could be linked to a location based on visual clues, while Raven targets photos taken indoors.