Paper2Video: Automatic Video Generation From Scientifuc Papers

📃Arxiv | 🌐 Project Page | 💻Github

Dataset Description

The Paper2Video Benchmark includes 101 curated paper–video pairs spanning diverse research topics. Each paper averages about 13.3K words, 44.7 figures, and 28.7 pages, providing rich multimodal long-document inputs. Presentations contain on average 16 slides and run for about 6 minutes 15 seconds, with some reaching up to 14 minutes. Rather than focusing only on video generation, Bench is designed to evaluate long-horizon agentic tasks that require integrating text, figures, slides, and spoken presentations.

Dataset Structure

This repository contains two main components:

• Excel file with metadata and presentation links
  Each entry includes:

  • paper: the title of the paper
  • paper_link: the URL of the paper (e.g., PDF or LaTeX source)
  • presentation_link: the URL of the author-recorded presentation video (some entries also include original slides)
  • conference: the conference where the paper was published
  • year: the publication year of the paper

• Author identity file
  This file contains author information, including voice samples and images, which can be used for tasks such as personalized talk synthesis or avatar generation.
  Each folder includes:

  • ref_img.png: the identity image of the author
  • ref_audio.wav: the identity voice sample of the author

Ethics

The author identity data (images and voice samples) provided in this repository are strictly for research purposes only. They must not be used for any commercial applications, deepfake creation, impersonation, or other misuse that could harm the rights, privacy, or reputation of the individuals. All usage should comply with ethical guidelines and respect the identity and intellectual property of the authors.

Citation

BibTeX:

@misc{paper2video,
      title={Paper2Video: Automatic Video Generation from Scientific Papers}, 
      author={Zeyu Zhu and Kevin Qinghong Lin and Mike Zheng Shou},
      year={2025},
      eprint={2510.05096},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2510.05096}, 
}