How 2R is a dataset for text-based video retrieval. The dataset contains 24,328 60-second clips and 51,390 related query terms collected from 9,371 videos in the HowTo 100M dataset, with an average of 2-3 related query terms per clip. 80% of the data is used for training, 10% of the data is used for verification, and 10% of the data is used for testing. How 2R and How 2QA are new challenging benchmarks that can be used to study the fields of video retrieval and video question answering.

Citation

@inproceedings{li2020hero,
title={HERO: Hierarchical Encoder for Video+ Language Omni-representation Pre-training},
author={Li, Linjie and Chen, Yen-Chun and Cheng, Yu and Gan, Zhe and Yu, Licheng and Liu, Jingjing},
booktitle={EMNLP},
year={2020}
}