Tumblr GIF (TGIF)

TGIF数据集是由罗切斯特大学和雅虎研究院的研究人员共同创建的，包含10万个从Tumblr收集的动画GIF及其通过众包获取的12万个自然语言描述。该数据集旨在为图像序列描述系统提供测试平台，特别是针对动画GIF或视频片段的自然语言描述生成。为了确保数据集的高质量，研究团队开发了一系列创新的质量控制措施，以验证众包工作者的自由文本输入。TGIF数据集的应用领域包括自动电影描述，旨在解决视频内容理解和描述生成的问题。

The TGIF Dataset was co-created by researchers from the University of Rochester and Yahoo Research. It contains 100,000 animated GIFs collected from Tumblr and 120,000 natural language descriptions obtained via crowdsourcing. This dataset is designed to serve as a testbed for image sequence description systems, particularly for natural language description generation targeting animated GIFs or video clips. To ensure the high quality of the dataset, the research team developed a series of innovative quality control measures to validate the free-text inputs from crowdworkers. Application scenarios of the TGIF Dataset include automatic movie description, which aims to address the challenges of video content understanding and description generation.