How mark-kitur/road_defects_detection works
This repository combines a Flask image-upload demo with a YOLO training and inference workflow. In the web path, a user submits an image; the app saves and resizes it, runs the loaded model, saves the annotated prediction, and renders a page that can display the result. Separately, road_defect.py trains a YOLO model and demonstrates prediction on test images. The README describes small/medium model comparison, but the sampled training script shows only a medium-model training setup; notebook internals and the checkpoint artifacts were not inspected.
Main entry points: app.py Flask app · road_defect.py YOLO training · upload.html Upload page · predicted_images Prediction files
Components 12
Web inference
- Flask app
app.py - Upload page
templates/upload.html - Image preparation
app.py - YOLO inference
app.py - Result serving
app.py
Model development
- YOLO training
road_defect.py - Training dataset
Model and results
- Model checkpoint
best_m.pt - Uploaded images
app.py - Prediction files
predicted_images
Other components
- Image user
- Ultralytics YOLO
Connections 14
- Image user → Flask app: submits image
app.py - Flask app → Upload page: renders
app.py - Flask app → Uploaded images: saves upload
app.py - Flask app → Image preparation: calls
app.py - Image preparation → Uploaded images: writes resized image
app.py - Flask app → YOLO inference: calls
app.py - YOLO inference → Model checkpoint: loads
app.py - YOLO inference → Ultralytics YOLO: predicts with
app.py - Flask app → Prediction files: saves prediction
app.py - Image user → Result serving: requests result
app.py - Result serving → Prediction files: serves file
app.py - Result serving → Upload page: renders result
app.py - YOLO training → Training dataset: trains on
road_defect.py - YOLO training → Ultralytics YOLO: trains and predicts with
road_defect.py