Products / VRAS Vision™

VRAS Vision™

Purpose built computer vision models tuned for infrastructure defect classes cracks, corrosion, thermal anomalies, solar hotspots, powerline vegetation, and more.

[V-VIS-01] · GA · v4.2Request Datasheet

Product · Console

Foundation vision models for infrastructure.

VRAS VISION™ · model.damage.tower.v4
INFERENCE
corrosion 0.94missing_bolt 0.82vegetation 0.71
frame 00184 · 3840×2160lat 184 ms · gpu 62%
Detections · frame 00184
corrosion0.94high
missing_bolt0.82med
vegetation0.71low
insulator_flash0.63low
Model
damage.tower.v4stable
classes14
precision0.93
recall0.91

Overview

What it does.

Vision™ is the model runtime that powers Insight™ and Sentinel™. Inference runs on vehicle for latency sensitive work and in cloud for batch backfill. Every detection carries confidence, provenance, and a Verify™-signed reference to the source frame.

Interfaces

API
detection-api · v1 · REST
SDK
TypeScript · Python · Rust (edge)
Runtime
PyTorch · TensorRT · ONNX
Models
damage.tower.v4 · corrosion.pipe.v2 · solar.hotspot.v3 …

Capabilities

Feature surface.

CAP-01

14 defect classes

Domain tuned models for power, pipeline, rail, roof, solar, and substations.

CAP-02

Edge + cloud

TensorRT optimized on vehicle; horizontally scaled GPU pools in cloud.

CAP-03

Custom training

Bring your own labels; adapt an existing base model in hours, not weeks.

CAP-04

Signed inference

Every detection references the exact frame hash and model version.

Performance envelope

0.94
Precision (best)
180 ms
Edge latency
8k fps
Cloud throughput
monthly
Model refresh

Evaluate

Bring VRAS Vision™ into your operations.

Datasheets, security packets, and reference architectures available under mutual NDA.