TNL2K
收藏资源简介:
TNL2K是一个专为自然语言引导跟踪设计的大型数据集,包含2000个视频序列,旨在为自然语言引导跟踪算法的发展和评估提供平台。数据集中的视频来自YouTube、智能监控摄像头和移动设备,每个视频都密集标注了目标对象的位置信息和英语句子描述。TNL2K特别关注表达目标对象的属性、类别、形状、特性以及与其他对象的结构关系,为跟踪提供丰富的细粒度外观信息和高层次语义信息。数据集分为1300个视频用于训练和700个视频用于评估,反映了对抗样本和模态切换等挑战,适用于评估域适应性和长期跟踪能力。
TNL2K is a large-scale dataset tailored for natural language-guided tracking, encompassing 2000 video sequences, and it serves as a dedicated platform for the development and evaluation of natural language-guided tracking algorithms. The videos in this dataset are sourced from YouTube, intelligent surveillance cameras, and mobile devices. Each video is densely annotated with the position information of target objects and English sentence descriptions. Specifically, TNL2K emphasizes capturing the attributes, categories, shapes, characteristics, and inter-object structural relationships of target objects, thereby providing abundant fine-grained appearance information and high-level semantic information for tracking tasks. The dataset is split into 1300 videos for training and 700 videos for evaluation. It incorporates challenges such as adversarial samples and modality switching, and is applicable for evaluating domain adaptation and long-term tracking capabilities.




