PV-VTT: A Privacy-Centric Dataset for Mission-Specific Anomaly Detection and Natural Language Interpretation

PV-VTT Dataset Overview

The PV-VTT (Privacy Violation Video To Text) dataset is a privacy-centric resource designed to advance research in privacy violation detection and video anomaly recognition. Unlike traditional datasets that focus on severe crimes, PV-VTT emphasizes precursor activities—such as trespassing or unauthorized surveillance—that may precede more serious offenses.

Key Features of PV-VTT:

• Multimodal Annotations: Each video in the dataset is accompanied by detailed textual descriptions, providing context and facilitating natural language interpretation.
• Privacy Preservation: To protect individual privacy, the dataset includes both video feature vectors, and raw video data. Using the former allows researchers to work with the data without compromising participant confidentiality.
• Diverse Scenarios: PV-VTT encompasses a wide range of privacy violation scenarios, offering a comprehensive resource for developing and testing detection models.

Terms of Use

By accessing the PV-VTT dataset, you agree to the following:

• Use the dataset solely for academic, non-commercial research purposes.
• Ensure the privacy and anonymity of individuals depicted in the dataset.
• Do not create derivative works that could identify individuals or private scenarios.
• Comply with all applicable laws and ethical guidelines.
• Acknowledge the PV-VTT dataset in your research publications.

Accessing or downloading any files implies acceptance of these terms.

Demo Videos

Below are sample videos demonstrating the dataset:

Download

Before downloading, please ensure you have read and agreed to the Terms of Use.

• Download Embedding Dataset and Annotations (PV-VTT original)

• Raw Video Dataset (Not in Paper)

Paper Citation (BibTeX)

Use the following BibTeX entry to cite this paper:

    @article{masukawa2024pv,
        title={PV-VTT: A Privacy-Centric Dataset for Mission-Specific Anomaly Detection and Natural Language Interpretation},
        author={Masukawa, Ryozo and Yun, Sanggeon and Yamaguchi, Yoshiki and Imani, Mohsen},
        journal={arXiv preprint arXiv:2410.22623},
        year={2024}
    }