MOET

我们在拥挤的现实世界视频中提供了一个公开的多对象眼动跟踪 (MOET) 数据集，该数据集由跟踪特定对象的参与者的凝视数据组成，并用标签和边界框注释，用于训练和评估注意力解码算法。我们还提出了两种用于注意力解码的端到端深度学习模型，并将其与最新的启发式方法进行了比较。We present a publicly available multi-object eye tracking (MOET) dataset sourced from crowded real-world videos. This dataset comprises gaze data collected from participants tracking specific objects, and is annotated with labels and bounding boxes, designed for training and evaluating attention decoding algorithms. We additionally propose two end-to-end deep learning models for attention decoding, and compare them against state-of-the-art heuristic methods.

搜集汇总

背景与挑战

背景概述

MOET是一个公开的多对象眼动跟踪数据集，包含现实世界视频中参与者的凝视数据，并带有标签和边界框注释，用于注意力解码算法的训练和评估。该数据集由卡内基梅隆大学等机构于2022年发布。