PV-VTT
收藏资源简介:
PV-VTT数据集由加利福尼亚大学欧文分校和芝浦工业大学创建,专注于隐私侵犯检测。该数据集包含702个视频,涵盖多种隐私侵犯场景,如偷窥、非法侵入等。数据集通过人工标注和大型语言模型(LLM)进行文本描述的匿名化处理,确保隐私保护。PV-VTT旨在通过视频异常识别和视频描述任务,提升机器学习模型在预防严重犯罪方面的能力。
The PV-VTT dataset was created by the University of California, Irvine and Shibaura Institute of Technology, focusing on privacy violation detection. This dataset contains 702 videos covering various privacy violation scenarios such as voyeurism, trespassing, and others. Anonymization processing for text descriptions is conducted via manual annotation and Large Language Models (LLMs) within the dataset to ensure privacy protection. PV-VTT aims to enhance the capabilities of machine learning models in preventing serious crimes through video anomaly recognition and video captioning tasks.

- 1PV-VTT: A Privacy-Centric Dataset for Mission-Specific Anomaly Detection and Natural Language Interpretation加利福尼亚大学欧文分校, 芝浦工业大学 · 2024年



