We present SCENE-pathy, a dataset and a set of baselines to study the visual selective attention (VSA) of people towards the 3D scene in which they are located. In practice, VSA allows to discover which parts of the scene are most attractive for an individual. Capturing VSA is of primary importance in the fields of marketing, retail management, surveillance, and many others. So far, VSA analysis focused on very simple scenarios: a mall shelf or a tiny room, usually with a single subject involved. Our dataset, instead, considers a multi-person and much more complex 3D scenario, specifically a high-tech fair showroom presenting machines of an Industry 4.0 production line, where 25 subjects have been captured for 2 min each when moving, observing the scene, and having social interactions. Also, the subjects filled out a questionnaire indicating which part of the scene was most interesting for them. Data acquisition was performed using Hololens 2 devices, which allowed us to get ground-truth data related to people’s tracklets and gaze trajectories. Our proposed baselines capture VSA from the mere RGB video data and a 3D scene model, providing interpretable 3D heatmaps. In total, there are more than 100K RGB frames with, for each person, the annotated 3D head positions and the 3D gaze vectors. The dataset is available here: https://intelligolabs.github.io/scene-pathy.
SCENE-pathy: Capturing the Visual Selective Attention of People Towards Scene Elements
Caputo A.;Farinella G. M.;
2023-01-01
Abstract
We present SCENE-pathy, a dataset and a set of baselines to study the visual selective attention (VSA) of people towards the 3D scene in which they are located. In practice, VSA allows to discover which parts of the scene are most attractive for an individual. Capturing VSA is of primary importance in the fields of marketing, retail management, surveillance, and many others. So far, VSA analysis focused on very simple scenarios: a mall shelf or a tiny room, usually with a single subject involved. Our dataset, instead, considers a multi-person and much more complex 3D scenario, specifically a high-tech fair showroom presenting machines of an Industry 4.0 production line, where 25 subjects have been captured for 2 min each when moving, observing the scene, and having social interactions. Also, the subjects filled out a questionnaire indicating which part of the scene was most interesting for them. Data acquisition was performed using Hololens 2 devices, which allowed us to get ground-truth data related to people’s tracklets and gaze trajectories. Our proposed baselines capture VSA from the mere RGB video data and a 3D scene model, providing interpretable 3D heatmaps. In total, there are more than 100K RGB frames with, for each person, the annotated 3D head positions and the 3D gaze vectors. The dataset is available here: https://intelligolabs.github.io/scene-pathy.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.