Clock and Calendar Dataset

Overview

This dataset comprises a small sample of two subsets: Clock and Calendar. Each subset contains images paired with corresponding questions and answers, making it suitable for evaluating the reasoning capabilities of multimodal large language models (MLLMs) for calendar and analog clock interpretation.

Dataset Structure

The dataset is organized into two main categories:

• Clock Dataset
• Calendar Dataset

Each entry in the dataset includes:

• type: Category of the clock/ year of calendar.
• question: A question related to the time or date displayed in the image.
• answer: The correct answer to the question.
• image: Image of clock or calendar.

Clock Dataset

The Clock dataset contains various types of clock images with associated questions and answers about the time shown.

Clock Data Categories:

• Basic clock: Standard analog clocks.
• Black dial clock: Clocks with a black dial background.
• No second hand: Clocks without a second hand.
• Easy: Clocks displaying simple hour-based times (e.g., 1-hour, 2-hour marks).
• Roman: Clocks using Roman numerals for hour representation.
• Hand as arrows: Clocks where the hands are represented as arrows.

Calendar Dataset

The Calendar dataset contains images of yearly calendars with questions related to specific dates.

Calendar Data Categories:

• Popular days: Questions related to Christmas and New Year's Day.
• Less popular day: Questions about the Ides of March (March 15th).
• Random date: Questions about a randomly selected date (November 21st).
• Count-based days: Questions about the 100th and 153rd day of the year.

Citation

@misc{saxena2025losttimeclockcalendar,
      title={Lost in Time: Clock and Calendar Understanding Challenges in Multimodal LLMs}, 
      author={Rohit Saxena and Aryo Pradipta Gema and Pasquale Minervini},
      year={2025},
      eprint={2502.05092},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2502.05092}, 
}

Acknowledgments

This dataset was compiled for research and development purposes.