Flickr30k Entities: Collecting Region-to-Phrase Correspondences for Richer Image-to-Sentence Models

The Flickr30k dataset has become a standard benchmark for sentence-based description. This paper presents Flickr30k Entities, which augments the158k captions from Flickr30k with 244k coreference chains, linking mentions of same entities across different captions for the same image, and associating with 276k manually annotated bounding boxes. Such annotations are for continued progress in automatic image description and grounded understanding. They enable us to define a new benchmark for of textual entity mentions in an image. We present a strong for this task that combines an image-text embedding, detectors for objects, a color classifier, and a bias towards selecting larger. While our baseline rivals in accuracy more complex state-of-the-art, we show that its gains cannot be easily parlayed into improvements on tasks as image-sentence retrieval, thus underlining the limitations of methods and the need for further research.