ChartAct: A Benchmark for Dynamic Chart Understanding
Paper • 2605.26994 • Published • 10
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
This repository is a preview release of ChartAct, an interactive benchmark for dynamic chart understanding. The full benchmark will be released after formal publication.
Paper: ChartAct: A Benchmark for Dynamic Chart Understanding
GitHub: https://github.com/wulin-wulin/OSWorld_Chart
data/dynamic_chart_preview.jsonl: 5 Dynamic Chart preview samples.data/dashboard_chart_preview.jsonl: 5 Dashboard Chart preview samples.evaluation_examples/: minimal OSWorld-style task configs for the preview samples.web/ and web_insert/: chart webpages required by the preview samples.label/ and question_answer/: metadata and QA annotations for the preview samples.This preview is intended to show the benchmark format and runnable framework without releasing the full evaluation set before publication.
from datasets import load_dataset
dataset = load_dataset("chartact-benchmark/ChartAct")
dynamic = dataset["dynamic_chart_preview"]
dashboard = dataset["dashboard_chart_preview"]
print(dynamic[0])
@misc{chartact2026,
title={ChartAct: A Benchmark for Dynamic Chart Understanding},
author={Muye Huang and Lin Wu and Lingling Zhang and Hang Yan and Zhiyuan Wang and Yumeng Fu and Zesheng Yang and Jun Liu},
year={2026},
eprint={2605.26994},
archivePrefix={arXiv},
primaryClass={cs.CV}
}