Per-class precision
Separate NMS per class prevents smoke plumes from suppressing fire flames or extinguisher detections in overlapping regions.
Innomium Vision · Edge YOLO
Fire & smoke detection
built for safety.
Ember is Innomium's cutting-edge YOLO model for fire, smoke, and fire extinguisher detection — very light, very strong, and tuned for warehouses, industrial sites, and mission-critical safety monitoring with 90% accuracy.
- Accuracy
- 90%
- Model size
- 9.8 MB
- Runtime
- Edge / Browser
Live detection
All 3 classes · Fire, smoke, extinguisher
Proven performance
From raw scene to classified hazards.
Ember detects fire, smoke, and fire extinguishers with per-class NMS, smoke merging, and color-prior filters — keeping alerts accurate even in complex industrial scenes.
Smoke merge & fire suppress
Fragmented smoke boxes merge into coherent plumes, while nested fire duplicates are suppressed to reduce false alerts.
Edge-ready ONNX
A compact ~9.8 MB model deploys on safety cameras, drones, and in-browser demos without a GPU farm.
Capabilities
Safety intelligence, without the overhead.
3 hazard classes
Detects fire, smoke, and fire extinguisher with class-aware remapping and per-class post-processing tuned for validator scoring.
Ultra-light ONNX
~9.8 MB footprint runs on CPU via ONNX Runtime Web in the browser or on edge hardware at the camera.
False-positive control
Color-prior filters, per-class hard NMS, and confidence thresholds reduce spurious detections in complex scenes.
Privacy-first demo
Inference runs entirely in your browser. Frames never leave your device — ideal for regulated safety and surveillance pilots.
Use cases
Where Ember delivers.
Warehouses & factories
Early fire and smoke detection in storage facilities, production floors, and loading bays.
Retail & offices
Monitor kitchens, server rooms, and common areas for fire hazards and smoke buildup.
Wildfire & perimeter
Detect flames and smoke plumes along forest edges, campsites, and remote perimeters.
Extinguisher compliance
Locate fire extinguishers in camera views for safety audits and equipment verification.
Interactive demo
Upload your own scene.
Run the production ONNX model in your browser — per-class NMS and color filters included.
No image loaded
Awaiting detection
Detection JSON
[]
Deploy Ember in your safety stack.
Very light. Very strong. 90% accuracy on hazard detection. Multi-class fire and smoke detection for warehouses, industrial sites, and edge deployments that need reliable alerts.