CxC stands for Crisscrossed Captions, which is an image caption dataset containing 247,315 manually labeled annotations. This dataset extends the development and testing scope of the MS-COCO dataset using semantic similarity ratings for image-text pairs, text-text pairs, and image-image pairs.

Citation

@article{parekh2020crisscrossed, title={Crisscrossed Captions: Extended Intramodal and Intermodal Semantic Similarity Judgments for MS-COCO}, author={Parekh, Zarana and Baldridge, Jason and Cer, Daniel and Waters, Austin and Yang, Yinfei}, journal={arXiv preprint arXiv:2004.15020}, year={2020} }