Name of the Dataset: Flickr30K-CFQ: Compact and Fragmented Query challenge dataset
Dataset Introduction: Flickr30K-CFQ was created to improve text-image retrieval research by modeling realistic user queries that are concise and fragmented (e.g., “family gathering” or “children bundled up”), unlike verbose captions in traditional datasets. It extends the original Flickr30K dataset with four new query types: Imagery Tags (305 abstract labels), Phrases (111,240 noun phrases), Triples (133,540 entity-relation pairs), and Fragments (139,607 multi-triple descriptions), totaling over three times the original corpus.
Download Link: https://www.dropbox.com/scl/fi/m80yf919dgezulht90k6n/Flickr30K-CFQ.zip?rlkey=5xzt8vat6tj1vdmtntmpkdeel&st=rf6ww7b1&dl=0
Relevant Paper: Haoyu Liu, Yaoxian Song, Xuwu Wang, Xiangru Zhu, Zhixu Li*, Wei Song*, Tiefeng Li “Flickr30K-CFQ: A Compact and Fragmented Query Dataset for Text-image Retrieval.”
