Lecture Video Visual Objects (LVVO) Dataset
收藏资源简介:
LVVO数据集是一个针对教育视频内容中视觉对象检测的新基准数据集。它由从245个讲座视频中提取的4000帧组成,涵盖了生物学、计算机科学和地球科学等学科。其中1000帧被手动标注为四个视觉类别:表格、图表-图形、照片图像和视觉插图。每个帧由两名标注者独立标注,并通过冲突解决过程由第三位专家审查和解决所有分歧。为了扩展数据集,采用半监督方法自动标注剩余的3000帧,形成LVVO 3k。完整的LVVO数据集为开发和评估教育视频中视觉内容检测的监督和半监督方法提供了宝贵的资源。
The LVVO dataset is a novel benchmark dataset for visual object detection in educational video content. It consists of 4,000 frames extracted from 245 lecture videos, covering disciplines such as biology, computer science, and earth science. Among them, 1,000 frames were manually annotated into four visual categories: tables, charts-graphs, photographic images, and visual illustrations. Each frame was independently annotated by two annotators, and all disagreements were reviewed and resolved by a third expert through a conflict resolution process. To expand the dataset, a semi-supervised approach was adopted to automatically annotate the remaining 3,000 frames, forming the LVVO 3k subset. The complete LVVO dataset provides a valuable resource for developing and evaluating both supervised and semi-supervised methods for visual content detection in educational videos.
LVVO数据集概述
数据集简介
- 名称:Lecture Video Visual Objects (LVVO) Dataset
- 用途:教育视频中的视觉对象检测基准
- 内容:包含讲座视频帧中的表格、图表、图像和插图等视觉内容的高质量标注
数据集详情
- 总图像数:4,000帧(从讲座视频中提取)
- 手动标注子集(LVVO 1k):1,000帧
- 自动标注子集(LVVO 3k):3,000帧
- 数据来源:videopoints.org的讲座录像
- 覆盖8位讲师、13门课程和3个领域(生物学、计算机科学、地球科学)
- 标注工具:VoTT by Microsoft
类别信息
| 类别ID | 名称 |
|---|---|
| 1 | Table |
| 2 | Chart-Graph |
| 3 | Photographic-image |
| 4 | Visual-illustration |
数据集版本
- 完整数据集:
LVVO 1k withCategories.zip:1,000张手动标注图像(含类别)LVVO 1k.zip:相同图像(单类别标注)LVVO 3k.zip:3,000张自动标注图像
- 样本数据集:包含10张标注图像的迷你版本
文件结构
plaintext
LVVO_x/
├── images/ # 所有.jpg图像
├── labels/ # 对应的.json标注文件
└── dataset_info.json # 元数据:类别名称、图像ID映射
标注格式
- JSON文件包含:
asset:图像元数据(文件名、ID、尺寸)objects:标注的视觉元素列表(类别ID和边界框坐标)
许可信息
- LVVO数据集:CC BY 4.0
- LDD衍生元数据:CC BY 4.0
- LPM衍生元数据:CC BY-NC-SA 4.0
引用格式
bibtex @article{biswas2025lvvo, title={Lecture Video Visual Objects (LVVO) Dataset: A Benchmark for Visual Object Detection in Educational Videos}, author={Dipayan Biswas and Shishir Shah and Jaspal Subhlok}, journal={arXiv preprint arXiv:2406.00123}, year={2025} }
获取方式
- 下载链接:https://drive.google.com/drive/folders/1RJjj71CAQHnRLqcWIYmJbWj_1zGmw2bO?usp=drive_link
- 论文链接:https://doi.org/10.48550/arXiv.2506.13657

- 1Lecture Video Visual Objects (LVVO) Dataset: A Benchmark for Visual Object Detection in Educational Videos德克萨斯大学休斯顿分校计算机科学系 · 2025年



