MolmoPoint-TrackAny

MolmoPoint-TrackAny 数据集包含来自 YouTube 和 MammalNet 的自然视频的点跟踪标注。每个样本包含一个描述对象的自然语言表达式、逐帧点轨迹以及链接回源视频的元数据。数据集包含 17,437 个唯一标注，覆盖 13,050 个视频，每个标注以 1 FPS 和 2 FPS 两种采样频率出现，共计 34,874 行。数据集的主要字段包括唯一标识符、视频名称、对象描述、原始视频 FPS、采样 FPS、视频高度和宽度、帧数、任务类型、逐帧点轨迹、可选掩码标识符、对象标识符、查询标识符、标注标识符、视频路径和视频来源。该数据集适用于视频分类、目标检测、视频对象跟踪和视频分割等任务。需要注意的是，数据集仅包含标注，源视频需单独下载。The MolmoPoint-TrackAny dataset contains point tracking annotations for natural videos sourced from YouTube and MammalNet. Each sample includes a natural language expression describing the object, frame-by-frame point trajectories, and metadata linking back to the original source video. The dataset comprises 17,437 unique annotations covering 13,050 videos, with each annotation available at two sampling frequencies: 1 FPS and 2 FPS, resulting in a total of 34,874 rows. The key fields of the dataset include unique identifier, video name, object description, original video FPS, sampling FPS, video height and width, total frame count, task type, frame-by-frame point trajectories, optional mask identifier, object identifier, query identifier, annotation identifier, video path, and video source. This dataset is applicable to tasks such as video classification, object detection, video object tracking, and video segmentation. Please note that the dataset only contains annotations, and the source videos need to be downloaded separately.