CivilSoftware
Crack Detection in Concrete
A CNN that flags cracks in photos of campus buildings for inspection.
The problem
Manual inspection of older blocks is slow and inconsistent.
What we built
Fine-tuned a small CNN on a public crack dataset plus 600 photos we labelled around campus.
Hardware
Phone camera on a telescopic pole.
Software
PyTorch, MobileNetV3, a Streamlit review tool.
Outcomes
94% precision on our held-out campus photos; the civil department now uses it for first-pass screening.
