metadata
license: cc-by-nc-sa-4.0
language:
- en
[Work in Progress] We release TimeR-v1, a dataset that aims to advance Video LLMs towards temporally grounded reasoning. We collect existing video-text datasets with timestamp annotations and construct a large-scale high-quality dataset with the help of powerful LLMs. Each instance in our dataset contains a question-answer pair, as well as a list of timestamps denoting the relevant video segments.