Visually-attentive UNREAL agents
收藏arXiv2025-09-30 收录
下载链接：
https://sites.google.com/view/attendbeforeyouact
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了用于在3D导航迷宫环境中训练视觉注意力UNREAL智能体的图像，其中包括原始图像和通过显著性地图生成的注视图像。数据集还包含了多种图像变体，如加入高斯噪声和染色效果的图像，以评估智能体在不同情境下的性能表现。该数据集已在不同次数的运行中进行了训练，平均跨越了7次运行，其任务是研究在导航任务中持续学习使用视觉注意力。
This dataset contains images used for training visual attention-augmented UNREAL AI Agents in a 3D navigation maze environment, including both raw images and gaze images generated from saliency maps. It also includes various image variants, such as those with added Gaussian noise and color tinting effects, to evaluate agent performance under different scenarios. This dataset has been utilized across multiple training runs, with an average of 7 runs per experimental setup, and the associated research goal is to investigate continual learning of visual attention utilization in navigation tasks.
提供机构：
DeepMind
