Google has released a new version of its Open Images dataset, which contains nine million annotated images. This new version includes synchronized audio, text, and mouse traces for the objects a narrator describes and hovers over for 500,000 images. The version also includes new text labels to describe visual relationships, such as “dog catching a flying disk.” This dataset can help advance AI systems’ ability to identify and determine the relationship between objects in visual data.
Image: Jean-noelkern