Flickr30K Entities Dataset

Flickr30K Entities is an extension of the Flickr30K dataset. Flickr30K Entities annotates the nouns mentioned in the description sentences of the Flickr30K dataset and annotates the corresponding bounding boxes.

The dataset adds 244K coreference chains to the original 158K description sentences, linking mentions of the same entity in different description sentences in the same image, and associates them with 276K manually annotated bounding boxes. The dataset is used to define a new benchmark for localizing textual entity mentions in images.