Visual Focus of Attention Dataset

Available downloads are:

Ground truth

The experiment scenario and recording conditions are demonstrated in the following video:

In our experiments, a human experimenter and a robotic agent sit opposite to each other and interact. As the experimenter switches his gaze on the objects in front of him, this scene is recorded using an on-board camera of the robotic agent. The ground truth of focus of attention of the experimenter is manually annotated.

We collected two sets of videos, comprising 20 sequences, recorded at a rate of 10-fps with 320 x 240 resolution. The whole database constitutes 6691 images in total.

Recording set-1 is designed to evaluate the intraclass and interclass generalization capabilities under normal illumination conditions, providing an understanding of the baseline performance, where each of the four experimenters provide two sequences in front of a stationary camera under sufficient natural illumination.

Recording set-2 is composed of 12 video sequences collected from four experimenters under three different conditions. These include artificial illumination, dimmed illumination, and motion blur.