ViVa-DataCreator: An Open-Source Human-in-the-Loop Data Annotation Engine for Semantic Segmentation

ViVa-DataCreator is an open-source tool for creating semantic segmentation datasets by tracking objects of interest from videos. It leverages the Segment Anything Model 2 (SAM2) and You Only Look Once (YOLO) AI models to perform segmentation and object detection, guiding users through an 8-step process to generate complete datasets ready for model training.

ViVa-DataCreator: Graphical User Interface for Dataset Creation

This tool focuses on generating semantic segmentation datasets through object tracking, utilizing SAM 2 to enhance segmentation accuracy.

Key Features

• Video-to-Dataset Conversion: Transform videos into high-quality segmentation datasets with minimal manual effort.
• SAM 2 Integration: Utilize the latest Segment Anything Model 2 for accurate and interactive segmentation.
• 8-Step Flexible Process: A comprehensive workflow that guides you through dataset creation, allowing you to move between steps as needed.
• Interactive Refinement: Manually refine segmentations for precision and quality control.
• Object Tracking Integration: Utilize YOLO and DeepSort for tracking objects of interest across video frames.
• Batch Processing: Efficiently handle large videos through configurable batch processing.
• Customizable Classes: Define and assign custom object classes with unique colors.
• Data Augmentation: Generate augmented datasets (random crops, rotations) to increase dataset size and variability.
• Safety-Focused: Designed for safe and reliable dataset generation without hardware risks.

Documentation

For detailed usage instructions, examples, and API documentation, please refer to the ViVa-DataCreator Documentation.

Citation

If you use ViVa-DataCreator in your research, please consider adding the following citations:

ViVa-DataCreator

@software{soriano2025datacreator,
  author = {Miguel Soriano-García, Diego Mercado-Ravell, Israel Becerra and Julio De La Torre-Vanegas},
  title = {ViVa-DataCreator: An Open-Source Human-in-the-Loop Data Annotation Engine for Semantic Segmentation},
  year = {2025},
  url = {https://github.com/viva-safeland/viva_datacreator}
}

ViVa-SAFELAND Simulator

@article{soriano2025viva,
  title={ViVa-SAFELAND: a New Freeware for Safe Validation of Vision-based Navigation in Aerial Vehicles},
  author={Miguel S. Soriano-Garcia and Diego A. Mercado-Ravell},
  journal={arXiv preprint arXiv:2503.14719},
  year={2024}
}

Related Application

@misc{delatorre2025riskaware,
      title={Vision-Based Risk Aware Emergency Landing for UAVs in Complex Urban Environments}, 
      author={Julio de la Torre-Vanegas and Miguel Soriano-Garcia and Israel Becerra and Diego Mercado-Ravell},
      year={2025},
      eprint={2505.20423},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2505.20423}, 
}