Paper2Video / README.md
nielsr's picture
nielsr HF Staff
Add task categories, relevant tags, paper link, and sample usage for Paper2Video dataset
a6eedd0 verified
|
raw
history blame
4.65 kB
metadata
license: mit
tags:
  - agent
  - multimodal
  - video-generation
task_categories:
  - any-to-any
language:
  - en

Paper2Video: Automatic Video Generation From Scientific Papers

📃 Paper | 🌐 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

Sample Usage

The PaperTalker framework, part of Paper2Video, enables automatic academic presentation video generation. Follow these steps to get started:

Requirements

Prepare the environment by installing necessary packages and cloning dependent repositories:

cd src
conda create -n p2v python=3.10
conda activate p2v
pip install -r requirements.txt

Download the dependent code and follow the instructions in Hallo2 to download the model weight.

git clone https://github.com/fudan-generative-vision/hallo2.git
git clone https://github.com/Paper2Poster/Paper2Poster.git

You need to prepare the environment separately for talking-head generation to potentially avoid package conflicts. Please refer to Hallo2. After installing, use which python to get the python environment path.

cd hallo2
conda create -n hallo python=3.10
conda activate hallo
pip install -r requirements.txt

Configure LLMs

Export your API credentials:

export GEMINI_API_KEY="your_gemini_key_here"
export OPENAI_API_KEY="your_openai_key_here"

Inference

The pipeline.py script provides an automated pipeline for generating academic presentation videos. It takes LaTeX paper sources together with reference image/audio as input to produce a complete presentation video. The minimum recommended GPU for running this pipeline is NVIDIA A6000 with 48G.

Run the following command to launch a full generation:

python pipeline.py \
    --model_name_t gpt-4.1 \
    --model_name_v gpt-4.1 \
    --model_name_talking hallo2 \
    --result_dir /path/to/output \
    --paper_latex_root /path/to/latex_proj \
    --ref_img /path/to/ref_img.png \
    --ref_audio /path/to/ref_audio.wav \
    --talking_head_env /path/to/hallo2_env \
    --gpu_list [0,1,2,3,4,5,6,7]

For a detailed list of arguments and further instructions, please refer to the GitHub repository.

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}, 
}