Infrastructure Inspection

Bridges, pipelines, power lines, and industrial structures need regular inspection. Inspectors climb scaffolding, rappel down structures, or review drone footage frame by frame. The process is slow, and subjective: two inspectors looking at the same crack may rate its severity differently.

Datature Vi trains a model on your inspection imagery. You label photos showing different types of damage (cracks, corrosion, spalling, deformation) at different severity levels, and the model learns to produce consistent assessments. When paired with drone footage, it can process hundreds of images from a single survey and flag the locations that need attention.

For an interactive overview of this application, visit the infrastructure inspection use case on vi.datature.com.

Common applications

Choose your task type

Annotation examples

Deploy and test

import json
from vi.inference import ViModel

model = ViModel(
    run_id="your-run-id",
    secret_key=".your-secret-key.",
    organization_id="your-organization-id",
)

result, error = model(
    source="drone_image.jpg",
    user_prompt="Assess the structural condition visible in this image.",
    generation_config={"temperature": 0.0, "do_sample": False}
)

if error is None:
    report = json.loads(result.result)
    print(f"Damage type: {report['damage_type']}")
    print(f"Severity: {report['severity']}")

Training tips

Use your own severity scale: define severity levels that match your organization's inspection standards (e.g., NBI condition ratings for bridges, API 570 for piping) and apply them consistently across all annotations.

Include clean/undamaged examples: the model needs to learn what a healthy structure looks like, not only damaged ones.

Capture from multiple distances: drone surveys produce images at different distances. Include close-ups (for detail) and wider shots (for context) in your training data.

Annotate with precise language: "hairline crack on the lower third" is more useful than "crack visible." Consistency and precision in annotations translate directly to consistency in model output.

Next steps