Welcome to the Cross-lingual-Test-Dataset-XTD10 corpus!

We introduce a new 1K multi-lingual MSCOCO2014 caption test dataset (XTD10) in 7 languages that we collected using a crowdsourcing platform. We use this as the test set for evaluating zero-shot model performance across languages. Please see paper for more detailed evaluation results. For more information on the MIC dataset (de, fr) please refer this and for STAIR dataset (jp) please refer this.

Licensing Information

Please see the Adobe License for this dataset here. Contact the authors for information on commercial use.

Citation

If you use the following languages from XTD-10 corpus in your publication, please cite this paper

• Italian, Spanish, Korean, Russian, Polish, Turkish, Chinese Simplified

@misc{aggarwal2020zeroshot,
      title={Towards Zero-shot Cross-lingual Image Retrieval},
      author={Pranav Aggarwal and Ajinkya Kale},
      year={2020},
      eprint={2012.05107},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Note

To read the extended version of this paper please see here. This paper was presented in the Workshop of Multilingual Search of The Web Conference, 2021.