TELL me what you See (TESE)

<p>Dataset reported in the deliverable D3.10</p> <p><strong>Description</strong></p> <p>TEll me what you SEe (TESE) dataset is a gaze annotated dataset, tailored for training deep learning architectures and extracting useful viewing patterns. It was created following the &ldquo;description&rdquo; paradigm where subjects were asked to orally describe visual scenes. Fifty-six different subjects participated in the recording procedure using Tobbi Pro eye tracker sensor. The publicly available Visual Genome (VG) dataset is a generic content image database with object bounding box, class and relation dense annotations.</p> <p><strong>Content</strong></p> <p>The gaze annotations are provided with the JSON file format using the following naming convention.<br> imageID.json<br> The imageID denotes the Visual Genome image id which is actually the filename of the image. Inside the JSON file gaze points annotations, timestamp and x,y coordinates and transcribed text from voice description are included as presented below.<br> {&ldquo;subjectID&rdquo;:1,<br> &ldquo;gaze&rdquo;:[{&ldquo;Time&rdquo;: 6105.218916422819, &ldquo;X&rdquo;: 0.5127838850021362, &ldquo;Y&rdquo;: 0.4649979770183563},<br> {&ldquo;Time&rdquo;: 6105.2274152988775, &ldquo;X&rdquo;: 0.507378339767456, &ldquo;Y&rdquo;: 0.4624013900756836}]<br> }<br> In addition, the subjects.json file provides information about the subjects as described below.<br> [{&ldquo;id&rdquo;: &ldquo;1&rdquo;, &ldquo;gender&rdquo;: &ldquo;Male&rdquo;, &ldquo;age&rdquo;: &ldquo;38&rdquo;},<br> {&ldquo;id&rdquo;: &ldquo;2&rdquo;, &ldquo;gender&rdquo;: &ldquo;Female&rdquo;, &ldquo;age&rdquo;: &ldquo;29&rdquo;}]</p>