Data2MV - A User Behaviour Dataset for Multi-View Scenarios

Identified as a necessity for SmoothMV, an head-tracking based multi-view solution developed at INESC TEC in Portugal, gaze data of users when visioning multi-view content was collected, processed and stored as dataset. The proposed dataset will be used as training data for future endeavours from SmoothMV, namely view prediction associated with user interests. Tracking data from head movement was obtained from 45 participants using an Intel Realsense F200 camera during viewing experiments with 7 different video playlists, each being viewed at least 17 times. This data set was made publicly available to the research community, contributing to reducing the current scarcity of such type of data. For storage optimization, saliency maps were subjected to image compression. Instructions for access to original saliency data are also provided. In addition, complementary tools for anyone to generate scatter plots and saliency maps from obtained data are provided.

本数据集是葡萄牙INESC TEC开发的基于头部追踪的多视图解决方案SmoothMV的必备支撑资源，其中收集、处理并存储了用户观看多视图内容时的视线数据。本数据集将作为SmoothMV后续研究的训练数据，具体用于关联用户兴趣的视图预测任务。本次实验共招募45名受试者，使用Intel Realsense F200摄像头采集头部运动追踪数据；实验中受试者共观看7组不同的视频播放列表，每组播放列表的观看次数均不少于17次。本数据集已面向科研社区公开，有助于缓解当前此类数据稀缺的现状。为优化存储，显著性图（saliency map）已进行图像压缩处理，同时提供了获取原始显著性数据的操作指南。此外，本数据集还附带了辅助工具，支持用户基于采集到的数据生成散点图与显著性图。