viva-datacreator 0.0.1

An open-source tool for creating image segmentation datasets from videos using SAM 2

ViVa-SAFELAND: Dataset Creation Tool

ViVa-SAFELAND is an open-source tool for creating semantic segmentation datasets by tracking objects of interest from videos. It leverages the SAM 2 (Segment Anything Model 2) and 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-SAFELAND: 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 Guided Process: Step-by-step workflow ensuring comprehensive dataset creation from frame extraction to final composition.
• 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.
• 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-SAFELAND in your research, please cite our work:

@software{soriano_garcia_viva_safeland_2025,
  author = {Miguel Soriano-García, Diego Mercado-Ravell, Israel Becerra and Julio De La Torre-Vanegas},
  title = {ViVa-DataCreator: Dataset Creation Tool},
  year = {2025},
  url = {https://github.com/viva-safeland/viva-datacreator}
}