About the Project
ASU (Augmented Segmentation Utility) is a desktop GUI application that integrates Meta AI’s Segment Anything Model (SAM) for semi-automatic image annotation and segmentation. The tool is designed to accelerate and standardize ground-truth generation for deep learning model training, particularly in medical and scientific imaging workflows.
The application has been demonstrated on MRI brain tumor segmentation, where it significantly reduced annotation time compared to fully manual methods, while maintaining high boundary accuracy.
Key Features
- SAM model integration with selectable variants (ViT-H and ViT-B) for different speed/accuracy trade-offs
- Interactive click-based prompting using positive and negative point inputs for refined segmentation control
- Project-based batch image management with New, Open, and Save Project workflows
- Multi-class labeling with fully customizable class names for any annotation taxonomy
- YOLO-format export for direct compatibility with downstream object detection model training
- CPU / GPU device selection with automatic fallback to CPU for accessibility across hardware
- Zoom and fit-to-window controls for precise pixel-level annotation work
- Undo / Redo and Delete annotation controls for an efficient correction workflow
Built With
Collaborators
This project is a tripartite collaboration spanning Indonesia, Malaysia, and Japan, bringing together expertise in photonics research, AI systems, and medical imaging.
Current Progress
The software has been registered for copyright protection in 2026. Development is ongoing, with the team currently working on expanding dataset support, integrating SAM-2 for video segmentation, and publishing findings from the MRI brain tumor segmentation case study.