Walking Awareness Dataset (WAD)
收藏资源简介:
Walking Awareness Dataset (WAD) 是由腾讯公司微信人工智能模式识别中心创建的一个多样化、广泛且无偏见的行走感知数据集,旨在为盲人行走任务提供公平的训练和测试基准。该数据集包含来自欧洲和亚洲的12000条视频-手动注释对,涵盖了多种场景和天气条件。数据集的创建过程包括从YouTube和实地录制视频,并通过详细的注释策略对视频进行场景和响应标注。WAD数据集的应用领域主要集中在利用视觉语言模型(VLM)为视障人士提供实时、简洁且信息丰富的行走提醒,旨在解决视障人士在行走过程中面临的挑战。
Walking Awareness Dataset (WAD) was developed by the WeChat Artificial Intelligence Pattern Recognition Center of Tencent Holdings Limited. It is a diverse, extensive and unbiased walking perception dataset, designed to serve as a fair training and testing benchmark for blind walking assistance tasks. The dataset contains 12,000 video-manual annotation pairs sourced from Europe and Asia, covering a wide range of scenarios and weather conditions. The construction of the WAD dataset involves collecting videos from YouTube and conducting on-site recordings, followed by detailed annotation strategies to label the scenes and corresponding responses of the videos. The main application scenarios of the WAD dataset focus on utilizing Visual Language Models (VLMs) to provide real-time, concise and informative walking reminders for visually impaired people, with the goal of addressing the challenges encountered by visually impaired individuals during their walking journeys.




