Examples v1.0.0 Early Access

Overview

• First public early-access release of ROCm-LS Examples and the Streamlit demo.
• Focused on ROCm-ready use-cases for MONAI and hipCIM.

New in this release

• MONAI: Training and inference for the spleen CT segmentation bundle
  • Training with dataset download, training-aligned transforms, validation, and optional checkpointing
• Pathology (MONAI + TorchVision + hipCIM)
  • Tile-based WSI inference CLI
  • CSV-driven patch workflows

Streamlit demo application

• Initial UI to run selected use-cases, visualize outputs (e.g., annotated overlays), and streamline setup
• Automatic device detection and simplified parameter entry for quick experimentation

Note

• Refer individual components for compatibility and requirements.

Known issues and limitations

• Training is single GPU/CPU
• Pathology background filtering uses a simple heuristic and may need tuning per dataset
• Streamlit UI is early-stage and may not expose all flags/modes yet