The upstream diagnose.py expects RGB input for YOLO detection. Revert to RGB since the detector needs 3 channels.
Breaking API changes in upstream: - TempFeatureExtractor no longer accepts model_path - Use FaultDiagnoser.diagnose() for detection + temp extraction - Image should be grayscale (matching file-path behavior)
The WEIGHTS_URL and WHL_URL variables are no longer needed since whl package and weights are now placed manually.
- Add FastAPI microservice for infrared equipment fault detection - Add dual pyproject.toml: local development and Docker deployment - Add MANUAL_SETUP.md for manual file placement instructions - Add Dockerfile for containerized deployment - Add sample .env.example for PROJECT_ROOT configuration - Configure gitignore for weights, lib, venv, and environment files