Multi-class precision
Separate NMS per class prevents cars from suppressing buses or trucks in overlapping traffic lanes.
Innomium Vision · Edge YOLO
Vantage is Innomium's cutting-edge YOLO model for multi-class vehicle detection — very light, very strong, and tuned for highways, intersections, and high-throughput traffic scenes with 93% accuracy.
Proven performance
Vantage detects cars, buses, trucks, and motorcycles with per-class NMS and TTA consensus — keeping counts accurate even in busy traffic.
Separate NMS per class prevents cars from suppressing buses or trucks in overlapping traffic lanes.
Horizontal-flip test-time augmentation merges views and boosts scores when both agree on a detection.
A compact ~19 MB model deploys on roadside cameras, toll systems, and in-browser demos without a GPU farm.
Capabilities
Detects bus, car, truck, and motorcycle with class-aware remapping and per-class post-processing tuned for validator scoring.
~19 MB footprint runs on CPU via ONNX Runtime Web in the browser or on edge hardware at the camera.
Box sanity filters, per-class hard NMS, and confidence thresholds reduce spurious detections in complex road scenes.
Inference runs entirely in your browser. Frames never leave your device — ideal for regulated traffic and surveillance pilots.
Use cases
Vehicle counts, lane occupancy, and flow analysis on highways and arterials.
Classify vehicle types at gates, toll booths, and busy city intersections.
Track trucks, buses, and cars across loading bays and fleet staging areas.
Power traffic analytics, congestion alerts, and multimodal counting pipelines.
Interactive demo
Run the production ONNX model in your browser — TTA and per-class NMS included.
[]
Very light. Very strong. 93% accuracy on the road. Multi-class vehicle detection for highways, toll systems, and edge deployments that need reliable counts by vehicle type.