Presentation Information
[3Yin-A-45]Gaze Generation Model Evaluation via Image Retrieval Tasks Using Gaze Behavior
〇Iwamoto Yuma1, Matsuno Shogo1 (1. The University of Electro-Communications)
Keywords:
gaze behavior,image retrieval evaluation,visual information acquisition
In order to replicate human gaze behavior, extensive research has been conducted on generating gaze data for image viewing. However, the evaluation of generated gaze data has primarily relied on metrics based on the consistency of spatial or temporal geometric structures. The perspective of how much useful information these data contain as a result of attentional allocation within an image has not been sufficiently explored. Previous studies have reported that image retrieval based on gaze data is possible to a certain extent, suggesting that observed gaze data contains features related to attentional allocation within images. Therefore, we hypothesized that the validity of gaze data generated by models could be evaluated through an image retrieval task. In this paper, we investigate the evaluation of generated gaze data through image retrieval based on the similarity of gaze distributions during image viewing.