October 3, 2025Open Access

Do Vision-Language Models See Visualizations Like Humans? Alignment in Chart Categorization

View Full Paper

Ask AI

Bookmark

Authors

PGPéter GyarmatiUniversity of Vienna MKManfred Klaffenböck LKLaura M. KoestenUniversity of Vienna

Discussion

Member takes

Implication

This analysis evaluates vision-language models’ alignment in visual categorization, indicating distinct perceptual differences in encoding and dimensionality.

Key Points

VLMs demonstrate accuracy in identifying chart purpose and dimensionality, yet exhibit difficulty with specific encoding types.
Expert labels from the VisType typology were utilized to assess the classification abilities of the VLMs in scientific visualizations.
The study involved 13 diverse VLMs focusing on classifying visual stimuli according to three specific criteria.
Findings indicate that larger VLMs do not automatically provide better performance, stressing the need for human oversight in visualization tasks.

Cite This Study

Gyarmati et al. (2025) studied this question.

synapsesocial.com/papers/68e02f3cf0e39f13e7fa24e0 https://doi.org/10.48550/arxiv.2509.05718

View Full Paper

Ask AI

Bookmark

Implication

This analysis evaluates vision-language models’ alignment in visual categorization, indicating distinct perceptual differences in encoding and dimensionality.

Do Vision-Language Models See Visualizations Like Humans? Alignment in Chart Categorization

Discussion

Member takes

Implication

Key Points

Cite This Study

Do Vision-Language Models See Visualizations Like Humans? Alignment in Chart Categorization

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider