Vision-cnc YOLO Segmentation Subsystem

CI-safe repository for CNC Vision segmentation, fallback detection, and local operator tooling.

CI mode

The default test run does not require internet access or runtime-heavy packages:

pytest -q

requirements.txt stays intentionally empty so CI can import modules and run pure-Python tests without numpy, cv2, torch, ultralytics, or PySide6.

Local runtime mode

For YOLO inference, OpenCV tools, and the PySide6 operator UI:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements-runtime.txt
python scripts/vision_detect_smoketest.py --image path/to/image.jpg
python scripts/vision_ui.py

Main components

• vision/detect/detector_api.py: detector contract, config loader, and hybrid orchestration.
• vision/detect/yolo_seg.py: YOLOv8-seg wrapper with lazy runtime imports.
• vision/detect/fallback_seg.py: CPU fallback segmentation path.
• vision/dataset/hard_cases.py: hard-case export for active-learning review.
• vision/ui/main_window.py: PySide6 operator UI for camera/image review.
• vision/calib/: camera intrinsics and plane calibration helpers.

Documentation

• docs/vision_yolo.md
• docs/calibration.md
• docs/training.md

Logging

Runtime logs are written to log.txt.