metadata

annotations_creators:
  - expert-generated
language:
  - en
language_creators:
  - machine-generated
  - found
license:
  - cc-by-4.0
multilinguality:
  - monolingual
pretty_name: ACL-Fig
size_categories:
  - 1K<n<10K
source_datasets:
  - original
tags: []
task_categories:
  - image-classification
task_ids:
  - multi-label-image-classification

Dataset Card for ACLFig Dataset

Dataset Description

Paper:
Leaderboard:

Dataset Summary

1758 total labelled images

The scientific figures dataset contains 1758 scientific figures extracted from 890 research papers(ACL). The scientific figures are in png format.

The dataset has been classified into 19 categories. These are

Algorithms
Architecture/Pipeline diagrams
Bar charts
Box Plots
Confusion Matrix
Graph
Line Chart
Maps
Natural Images
Neural Networks
NLP rules/grammar
Pie chart
Scatter Plot
Screenshots
Tables
Trees
Pareto chart
Venn Diagram
Word Cloud

The scientific figures are in the png directory.

The metadata directory contains metadata extracted from the pdf along with scientific figures in json format.

Finally, the scientific_figures.csv file contains following columns/fields:

sci_fig : Scientific figure name
caption: Caption text
inline_reference: Scientific figure contexts mentioned in the research paper
metadata: metadata json filename
label: One of the 19 categories as described above.
acl_paper_id: Unique identifier assigned to each pdf by ACL

Supported Tasks and Leaderboards

Multi-label classification

Dataset Creation

The dataset was created using papers in ACL Anthology.

Annotations

Annotation process

~2k images manually labelled

Citation Information

TODO

Contributions

Thanks to @zebaKarishma, @shauryr and @KavyaPuranik for adding this dataset.