PMC-Vid: A Large-Scale Biomedical Video Captioning Dataset

77K biomedical video–caption pairs curated from PubMed Central Open Access (PMC-OA) papers.
The PMC-Vid was presented at the ICCV 2025 CVAMD Workshop; the paper is available on the CVF Open Access website.

---

Dataset Files

File	Description
pmc-vid_paper.csv	Paper-level metadata, including download paths and journal information
pmc-vid_caption.csv	Video–caption pairs (77K entries)
download_pmc.py	Helper script to download video files from PMC

---

Dataset Structure

pmc-vid_paper.csv

Paper-level metadata (merged from PMC Open Access file list and curated paper data). oa_file_list.csv is available on PMC server.

Column	Description
PMC_ID	Unique PMC article identifier
PMID	PubMed ID
title	Paper title
abstract	Abstract text
journal	Journal name
main_category	Main category
sub_category	Subcategory of the paper
File	Path to the PMC archive package (from oa_file_list.csv)
Article Citation	Citation text (journal, volume, pages, etc.) (from oa_file_list.csv)
Last Updated (YYYY-MM-DD HH:MM:SS)	Metadata update time (from oa_file_list.csv)
License	License type (from oa_file_list.csv)

---

pmc-vid_caption.csv

Video-level metadata and generated captions.

Column	Description
PMC_ID	Corresponding paper ID
mp4_name	Video filename
nxml_file	Associated NXML file name
caption	Caption text
category	Video category
duration	Video duration (seconds)
fps	Frames per second
width, height	Video resolution
file_size	File size in bytes
frame_count	Total frame count
aspect_ratio	Aspect ratio
bitrate	Bitrate (kbps)
data	Internal reference metadata

---

Downloading Videos

We provide a helper script to download the original videos directly from PMC:

# Download videos using paper information
python download_pmc.py pmc-vid_paper.csv ./videos --workers 8

This script uses the official PMC Open Access FTP structure for reproducibility.

---

License

The PMC-Vid dataset is released under the Creative Commons Attribution-NonCommercial (CC BY-NC 4.0) license. You are free to use, share, and adapt the dataset for non-commercial research and educational purposes, provided that proper attribution is given.

The generated captions were created using OpenAI GPT models. Accordingly, users must also comply with the OpenAI Terms of Use and Sharing & Publication Policy.

Note: Each individual video in the dataset originates from PMC and retains its original license. When using or redistributing individual videos, please follow the specific license terms provided by the source article on PMC.

---

Citation

If you use PMC-Vid in your research, please cite:

@InProceedings{Yamagishi_2025_ICCV,
    author    = {Yamagishi, Yosuke and Saito, Kuniaki and Hashimoto, Atsushi and Ushiku, Yoshitaka},
    title     = {PMC-Vid: A Large-Scale Biomedical Video Captioning Dataset},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops},
    month     = {October},
    year      = {2025},
    pages     = {1254-1264}
}