Flickr30k Entities: Collecting Region-to-Phrase Correspondences for Richer Image-to-Sentence Models
This provides a richer dataset for image description and grounded language understanding, though it is incremental as it builds on an existing standard benchmark.
The authors created Flickr30k Entities, a dataset that augments Flickr30k with 244k coreference chains and 276k bounding boxes to link textual entities to image regions, enabling a new benchmark for entity localization. They developed a baseline model that rivals complex state-of-the-art methods in accuracy but shows limited gains in tasks like image-sentence retrieval, highlighting current method limitations.
The Flickr30k dataset has become a standard benchmark for sentence-based image description. This paper presents Flickr30k Entities, which augments the 158k captions from Flickr30k with 244k coreference chains, linking mentions of the same entities across different captions for the same image, and associating them with 276k manually annotated bounding boxes. Such annotations are essential for continued progress in automatic image description and grounded language understanding. They enable us to define a new benchmark for localization of textual entity mentions in an image. We present a strong baseline for this task that combines an image-text embedding, detectors for common objects, a color classifier, and a bias towards selecting larger objects. While our baseline rivals in accuracy more complex state-of-the-art models, we show that its gains cannot be easily parlayed into improvements on such tasks as image-sentence retrieval, thus underlining the limitations of current methods and the need for further research.