◆Scientific LLM Benchmarks
GitHub
← All benchmarks
General· scientific-chart-reasoning

CharXiv

Princeton University / University of Wisconsin-Madison / HKU · 2024

2,323 expert-curated charts from scientific papers with descriptive and reasoning questions for realistic multimodal chart understanding.

GitHub stars
Task type
VQA
Modality
multimodal
Access
open
Size
2,323 items
License
CC-BY-SA-4.0
Metrics
accuracy
image
<image/binary>
category
cs
year
20
original_figure_path
arXiv_src_2002_041/2002/2002.12459/experiments/ssjoptimizations.jpg
original_id
2002.12459
figure_path
images/1.jpg
num_subplots
1
subplot_row
0
subplot_col
0
descriptive_q1
13
descriptive_q2
7
descriptive_q3
18
descriptive_q4
14
reasoning_q
Among all bars, what is the least time percentage difference between any two neighboring bars?
reasoning_q_source
2
reasoning_a_type
4
image
<image/binary>
category
cs
year
20
original_figure_path
arXiv_src_2006_021/2006/2006.05993/figs/boundary_hueristic_plot.jpg
original_id
2006.05993
figure_path
images/4.jpg
num_subplots
1
subplot_row
0
subplot_col
0
descriptive_q1
3
descriptive_q2
2
descriptive_q3
13
descriptive_q4
15
reasoning_q
Is the number of red points more than blue points?
reasoning_q_source
3
reasoning_a_type
2

Real rows from the Hugging Face datasets server · long values truncated