Sketch-DAVIS16, Sketch-DAVIS17, Sketch-YouTube-VOS

本研究介绍了三个基于草图的视频对象分割数据集：Sketch-DAVIS16, Sketch-DAVIS17, Sketch-YouTube-VOS，这些数据集利用人类手绘草图作为视频对象分割的低成本且信息丰富的参考。数据集扩展自DAVIS16, DAVIS17和YouTube-VOS，共包含765条草图标注。创建过程中，参与者在未观看完整视频的情况下，根据第一帧中的目标对象进行草图绘制，确保草图的多样性和实用性。这些数据集主要用于视频对象分割任务，旨在通过草图参考提高分割算法的效率和准确性，解决传统基于照片掩码或语言表达的分割方法的局限性。

This study introduces three sketch-based video object segmentation datasets: Sketch-DAVIS16, Sketch-DAVIS17, and Sketch-YouTube-VOS, which utilize human-drawn sketches as low-cost and informative references for video object segmentation. These datasets are extended from DAVIS16, DAVIS17, and YouTube-VOS, containing a total of 765 sketch annotations. During the dataset creation process, participants drew sketches based on the target objects in the first frame without watching the full video, ensuring the diversity and practicality of the sketches. These datasets are primarily used for video object segmentation tasks, aiming to improve the efficiency and accuracy of segmentation algorithms via sketch-based references and address the limitations of traditional segmentation methods based on photo masks or linguistic descriptions.