Affective VLM Leaderboard
Can VLMs predict human emotional responses to images?
None defined yet.
The Affective Computing Group was established by the Center for Cognitive Neuroscience and the Center for Artificial Intelligence at Adam Mickiewicz University in Poznań. We study human emotions using modern technologies. We want emotion research to help build better and more responsible technologies, and we want technology to help us study emotions better. To understand human emotions, we combine AI tools, data analysis and knowledge from affective science.
Affective VLM Leaderboard: how well do vision-language models predict human emotional responses to images? Models are evaluated zero-shot on IAPS, NAPS and LAI-GAI, on emotion classification and on prediction of Likert ratings, and compared against a human inter-rater ceiling.
LAI-GAI: Library of AI-Generated Affective Images. 847 AI-generated images designed to evoke 12 emotions. Each image has human ratings on 18 discrete and dimensional affective scales. The full library, with prompts and metadata, is at affectdatabases.amu.edu.pl.