Skip to content

Request for more details on PresentQuiz and IP Memory evaluation data #24

@snooow1029

Description

@snooow1029

Hi authors,
Thanks for sharing this fantastic work and open-sourcing the code! I am very interested in your evaluation metrics, specifically PresentQuiz and IP Memory.
I noticed the evaluation scripts are available in the repository (e.g., under src/evaluation/PresentQuiz and IPMemory), and the main benchmark dataset is hosted on Hugging Face (ZaynZhu/Paper2Video).
Could you please provide a bit more detail on the following?

  1. PresentQuiz Questions: Are the exact multiple-choice questions (QA sets) used for the 101 papers available in the Hugging Face dataset? If so, could you point me to the specific JSON/file format?

  2. IP Memory Data: For the IP memory evaluation, how are the 5-second video clips and the corresponding queries structured in the dataset?

If there are any specific system prompts or intermediate generated data (like the generated question pairs from construct.py) that you could share as a quick example, it would be extremely helpful for reproducing the evaluation!
Thanks again for the great contribution to the community!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions