LiveFood
收藏资源简介:
LiveFood数据集是由字节跳动公司收集的高质量美食视频数据集,包含超过5100个视频,总时长197小时,分为四个精细标注的领域:食材、烹饪、展示和食用。该数据集旨在支持增量视频亮点检测任务,特别是在实际应用中,关注的亮点领域和训练数据随时间增加。数据集的创建过程涉及从在线视频中精选内容,并通过专业人员进行质量控制和精细标注。LiveFood数据集适用于视频亮点检测和领域增量学习任务,旨在解决现有方法在处理不断增加的亮点领域和数据时的局限性。
The LiveFood dataset is a high-quality food video dataset collected by ByteDance. It contains over 5,100 videos with a total duration of 197 hours, and is divided into four finely annotated domains: ingredients, cooking, display, and consumption. This dataset aims to support incremental video highlight detection tasks, especially in real-world applications where the target highlight domains and training data increase over time. The creation of the dataset involves selecting content from online videos, followed by quality control and fine annotation conducted by professionals. The LiveFood dataset is applicable to video highlight detection and domain incremental learning tasks, and is designed to address the limitations of existing methods when handling the growing number of highlight domains and associated data.



