What type of study is this?

October 10, 2025Open Access

Can Vision Language Models Infer Human Gaze Direction? A Controlled Study

View Full Paper

Ask AI

Bookmark

Authors

Controlled study measures gaze direction inference in vision language models and humans, suggesting limited interaction capabilities.

Only 17% of vision language models accurately infer gaze direction, while humans achieve near-ceiling accuracy.
Among 111 VLMs assessed, 94 performed no better than random guessing, indicating a significant limitation.
The study utilized photos with varying difficulty to test gaze inference performance across different task conditions.
Findings suggest VLMs are not yet effective for natural human interaction, although they show potential for improvement.

Zhang et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark