How polisetti-bharath/roadshield works
RoadShield is a Streamlit road-damage inspection app: users choose live camera, image, or video input; YOLOv8 detects and labels damage, and image/video pothole detections can be turned into location-aware reports. The repository also includes a dataset-preparation, model-training, and evaluation workflow. Page-to-helper links and image/video inference are visible in excerpts; the realtime page and notebook internals were not sampled, so their details are limited to documented capabilities.
Main entry points: report.py Report helpers · 2_Image Detection.py Image detection · 3_Video Detection.py Video detection · YOLOv8_Small_RDD.pt YOLOv8 weights
Components 16
Inspection app
- Home screen
Home.py - Live detection
pages/1_Realtime Detection.py - Image detection
pages/2_Image Detection.py - Video detection
pages/3_Video Detection.py - YOLOv8 weights
models/YOLOv8_Small_RDD.pt - Shared UI
sample_utils/ui.py
Damage reporting
- Report helpers
sample_utils/report.py - Saved contacts
- Browser geolocation
- Nominatim
- Authority channels
Model development
- CRDDC2022 data
- Dataset preparation
training/0_PrepareDatasetYOLOv8.ipynb - Model training
training/1_TrainingYOLOv8.ipynb - Evaluation
training/2_EvaluationTesting.ipynb
Other components
- Inspector
Connections 19
- Inspector → Home screen: opens app
README.md - Inspector → Live detection: chooses live mode
- Inspector → Image detection: chooses image mode
- Inspector → Video detection: chooses video mode
- Home screen → Shared UI: renders with
Home.py - Image detection → Shared UI: renders with
pages/2_Image Detection.py - Video detection → Shared UI: renders with
pages/3_Video Detection.py - Image detection → YOLOv8 weights: loads model
pages/2_Image Detection.py - Video detection → YOLOv8 weights: loads model
pages/3_Video Detection.py - Live detection → YOLOv8 weights: uses model
- Image detection → Report helpers: calls reporting
pages/2_Image Detection.py - Video detection → Report helpers: calls reporting
pages/3_Video Detection.py - Report helpers → Saved contacts: reads and writes
sample_utils/report.py - Report helpers → Browser geolocation: requests location
sample_utils/report.py - Report helpers → Nominatim: reverse geocodes
sample_utils/report.py - Report helpers → Authority channels: prepares report channels
sample_utils/report.py - CRDDC2022 data → Dataset preparation: provides dataset
README.md - Dataset preparation → Model training: prepares data for
README.md - Model training → Evaluation: model evaluated by
README.md