xiaozhang79/CamTrack-53K

CCTV-Anomaly

📑 Paper | 💻 Code | 🤗 Model & Datasets

✨ Dataset Description

CamTrack-53K is a large-scale, diverse, and high-quality training dataset for camera viewpoint control. It contains 53K viewpoint trajectory annotations, organized into SFT and RL training subsets. 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

CamTrack-53K contains a total of 53,059 viewpoint trajectory annotations. The corresponding videos are sourced from:

• MeViS (8,459 annotations)
• A2D (26,221 annotations)
• YouTube-VOS (18,379 annotations)

🚀 Usage

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

📝 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}
}