tejas-varshney/duoqueue-civic-hazard-detector

DuoQueue Civic Hazard Detector

An interactive Hugging Face Space for the DuoQueue YOLOv8s model trained to detect three civic infrastructure hazards: pothole, waterlogged_road, and drain_overflow.

The Space provides image and video inference. The default confidence threshold is 0.30, and video inference processes every fourth frame to match the documented CPU-oriented pipeline.

Model

The model weights are provided in best.pt. The model uses the Ultralytics YOLO format and runs on CPU by default for broad Hugging Face Spaces compatibility.

Dashboard integration

The deployed Space can be used directly as a browser dashboard. For programmatic integration, use the Gradio client against the Space URL after deployment:

from gradio_client import Client, handle_file

client = Client("YOUR_USERNAME/duoqueue-civic-hazard-detector")
result = client.predict(
    handle_file("sample.jpg"),
    0.30,
    api_name="/predict"
)
print(result)

The exact API route can be confirmed from the Space's API page after the build completes.

Local development

pip install -r requirements.txt
python app.py