Repository navigation
Summary
This MR completes a major refactor of SAMM. A streamlined GUI, more robust service management, automated setup, and continuous integration and testing. Model inference and fine-tuning are fully manageable from the 3D Slicer GUI. Users no longer need to launch or manage the inference server from the command line.
Key Changes
Slicer-managed service
- Added automatic service setup, startup, connection, monitoring, logging, and shutdown.
- Removed the need to launch the inference server manually.
- Added autosaving, heartbeat monitoring, and improved recovery behavior.
Expanded model support
Added support for all available weights and their compatible prompt types across:
- SAM 1
- SAM 2.1
- MobileSAM
- MedSAM
- MedSAM Text
- MedSAM2
- SAM 3
- Medical-SAM3
- FastSAM
Fine-tuning
- Added a MedSAM2-style fine-tuning workflow.
- Integrated dataset preparation, training, evaluation, progress monitoring, and trained-weight discovery into Slicer.
Improved user experience
- Streamlined the GUI and prediction workflows.
- Added clearer controls, tooltips, status information, progress reporting, cancellation, and error handling.
- Improved segmentation, embedding, model, and job management.
Setup and testing
- Automated model-variant setup and checkpoint preparation.
- Added continuous integration and automated testing.