CCTV-Anomaly

📑 Paper | 💻 Code | 🤗 Model & Datasets

✨ Dataset Description

CCTV-Anomaly is a large-scale, diverse, and high-quality training dataset for surveillance video understanding. It contains 14,133 videos across 10 anomaly categories, with detailed captions and event annotations. It was proposed in our paper Thinking with Cameras: Active Visual Reasoning via Dynamic Viewpoint Control for Surveillance Video Understanding and used for training the CamVLM model.

📊 Dataset Statistics

• Total Videos: 14,133
• Total Category Labels: 14,133
• Total Event Descriptions: 14,133
• Total Video Duration: 234 h

🚀 Usage

To download and use the dataset for training, please refer to the instructions in our GitHub Repository.

Terms of Use

1. CCTV-Anomaly dataset is ONLY for research and non-commercial use. The annotations of CCTV-Anomaly dataset belong to SenseTime Research, and are licensed under a Creative Commons Attribution-NonCommercial 4.0 License.

2. For commercial usage of our CCTV-Anomaly annotations, please contact Mr. Malon (machang[at]tetras[dot]ai) and cc Sheng Jin (jinsheng13[at]foxmail[dot]com).

📝 Citation

If you find our work helpful for your research and applications, please cite our paper:

@article{zhang2026thinking,
  title={Thinking with Cameras: Active Visual Reasoning via Dynamic Viewpoint Control for Surveillance Video Understanding},
  author={Zhang, Xiao and Zeng, Wang and Jin, Sheng and Liu, Wentao and Qian, Chen and Kan, Shichao},
  journal={arXiv preprint arXiv:2609.06475},
  year={2026}
}