AI-powered concrete surface inspection system. Detects cracks, measures physical width in millimeters using distance transform medial axis, classifies severity (EN 206), and saves GPS-tagged logs. Built with FastAPI, ONNX Runtime (YOLOv8), Vanilla JS, and Capacitor.js.

This project uses a Streamlit-based interface for crack detection and analysis.

Label-efficient bridge crack classification on SDNET2018 — benchmarking DINO, MoCo v3, MAE vs supervised ViT pretraining under 5–100 % labels.