Yale-BIDS-Chen/medpmc-multi-fig-detection-dataset

MedPMC Multipanel Figure Detection Training/Test Datasets

Overview

This dataset contains the annotations used for the multipanel figure detection stage of the MedPMC framework. The task is to identify whether a biomedical figure is a multipanel, or compound, figure. This stage supports downstream subfigure extraction and caption alignment by separating figures that contain multiple visual panels from standalone figures. The training and validation sets combine GPT-4o–annotated PMC figures (source == "PMC_synthetic") and data derived from the ImageCLEF 2016 medical task (source == "ImageCLEF"). The test set consists of 1,615 ImageCLEF samples and 500 manually annotated PMC figures. For details on dataset construction, annotation guidelines, and the overall MedPMC pipeline, please refer to our paper.

Dataset Statistics

Split	Samples
Train	56,135
Validation	13,991
Test	3,956

Dataset Features

Field	Description
image	Figure image
image_id	Unique image identifier
caption	Figure caption
is_multipanel	Binary label indicating whether the figure is a multipanel figure
source	Source dataset or annotation subset from which the sample was derived

Label Definition

A figure is labeled as is_multipanel = true if it contains multiple visually distinct panels, subfigures, or components within a single figure image. These figures often include panel labels such as A, B, C, or D, or otherwise present multiple image regions, plots, diagrams, or visual elements intended to be interpreted as separate panels.

A figure is labeled as is_multipanel = false if it is a standalone figure without clearly separated visual panels.

Quick Start

Load the Dataset

from datasets import load_dataset

dataset = load_dataset("Yale-BIDS-Chen/medpmc-multi-fig-detection-dataset")

print(dataset)

Citation

If you use this dataset, please cite:

Citation information will be updated soon.

Part of this dataset is derived from the ImageCLEF 2016 medical task dataset. Please also cite the original ImageCLEF dataset and associated publication:

@InProceedings{GSB2016,

Title = {Overview of the {ImageCLEF} 2016 medical task},
Author = {Garc\'ia Seco de Herrera, Alba and Schaer, Roger and Bromuri, Stefano and M\"uller, Henning },
Booktitle = {Working Notes of {CLEF} 2016 (Cross Language Evaluation Forum)},
Year = {2016},
Month = {September},
Location = {\'Evora, Portugal}
}

License and usage

This dataset is distributed under CC BY-NC-SA 4.0 for non-commercial research use. Users are responsible for complying with the licenses and terms.

Questions?

For questions or feedback, please contact Hyunjae Kim at hyunjae.kim@yale.edu.