smpas 0.1.2

AprilTag-calibrated dried shiitake mushroom cap analysis (diameter, crack ratio, pattern classification).

SMPAS (Shiitake Mushroom Pattern Analysis System)

SMPAS is an AprilTag-calibrated analysis system for dried shiitake mushroom caps. It measures real-world diameter, estimates crack/pattern ratio, and classifies pattern type using a YOLO detector and SAM-based segmentation. It ships with a web UI for image upload and visualization.

Key Features

• AprilTag-based calibration for real-world measurements
• Diameter measurement (mm)
• Crack/pattern ratio estimation
• Pattern type classification
• Web app for interactive upload and results
• CLI tools for model download and web startup

System Requirements

• Python 3.10+
• CUDA GPU recommended for speed (CPU is supported)
• Network access to download models

Install

pip install smpas

SAM must be installed separately (PyPI does not allow direct Git dependencies):

pip install git+https://github.com/facebookresearch/segment-anything.git

Download Models

smpas-download-models

This downloads:

• yolo_mushroom.pt (default URL hosted by you)
• sam_vit_b.pth (official SAM checkpoint)

To override URLs:

smpas-download-models --yolo-url <URL> --sam-url <URL>

To download only one model:

smpas-download-models --no-yolo
smpas-download-models --no-sam

Run the Web App

smpas-web

Default address:

http://localhost:5000

Default Directories SMPAS creates these under ~/.smpas automatically:

~/.smpas/
├── models/      # model weights
├── data/        # optional input images
├── uploads/     # web uploads
└── outputs/     # optional outputs

Environment Variables

• SMPAS_HOME: base directory (default ~/.smpas)
• SMPAS_MODELS_DIR: model directory override
• SMPAS_DATA_DIR: data directory override
• SMPAS_UPLOAD_DIR: upload directory override
• SMPAS_DEVICE: cuda or cpu
• SMPAS_YOLO_URL: default YOLO download URL for smpas-download-models
• SMPAS_SAM_URL: default SAM download URL for smpas-download-models
• SMPAS_YOLO_MODEL: explicit YOLO model path
• SMPAS_SAM_MODEL: explicit SAM model path
• SMPAS_APRILTAG_SIZE: AprilTag size in mm (default 37.58)

CLI Options (smpas-web)

• --host, --port: bind address
• --base-dir: base directory (overrides SMPAS_HOME)
• --models-dir: model directory
• --data-dir: data directory
• --uploads-dir: upload directory
• --device: cuda or cpu
• --apriltag-size: AprilTag size in mm
• --max-workers: parallel workers
• --no-parallel: disable parallel processing

Data Requirements

• Images should include an AprilTag (tag16h5, ID 25)
• AprilTag black square size: 37.58 mm
• Prefer top-down images with minimal perspective distortion
• Recommended resolution >= 2000×2000

Output (Web Response) The web API returns:

• AprilTag detection status
• px/mm calibration
• Diameter (mm)
• Crack/pattern ratio (%)
• Pattern type classification
• Flatness score and suspected “ban-gu” flag

Notes

• Model weights are not bundled with the package.
• SAM must be installed manually from GitHub.
• GPU is optional but strongly recommended.

License Proprietary. See LICENSE.txt.