Sky-OG-Model — PV Thermal Anomaly Detection & Module Segmentation

Models for the Sky-OG PV thermal inspection pipeline.

Status: scaffold only. Weights and finalized configs are produced in RunPod and pushed here. The tree below is the intended layout.

Repository structure

Sky-OG-Model/
├── README.md
├── detection/
│   ├── yolo26_pv_4class_base.pt    # YOLO26 detection, 4-class (base)
│   └── yolo26_pv_4class_ft.pt      # + fine-tuned on Sky-OG frames
├── segmentation/
│   └── yolo11seg_module_autogeo.pt # YOLO11-seg, module auto-geo labels
└── configs/
    ├── data_detection.yaml
    ├── data_segmentation.yaml
    └── sahi_inference.yaml

Evaluation

Metrics vs. Sitemark ground truth and per-class confusion matrix to be added once training completes.