Smoke detection: Datasets & Models

Manual monitoring of camera walls for early signs of fire is error-prone: operators miss faint plumes, get fatigued, and can only watch so many feeds at once. Traditional point smoke detectors are useless for large open spaces, perimeters, warehouses, and wildland areas where surveillance cameras already exist. Automating smoke detection with segmentation lets you catch a plume in its first seconds across dozens of feeds simultaneously and localize exactly where it is originating. That early, spatially precise alert buys critical response time before a small incident becomes a major loss.

Segmentation masks give you not just a yes/no alert but the size, shape, and spread rate of a plume, reducing false alarms from steam or dust. The model runs on the RTSP and IP camera infrastructure you already have, so there is no need to install dedicated fire-detection hardware across every zone. Because it is trained on a labeled smoke dataset, you can expand coverage to new sites and camera types by adding representative frames and retraining.

1. Upload an image: Drop any photo into the widget above to try smoke detection for free, right on this page.
2. Preview instant results: The model detects smoke, smoke plume, fire, flame, steam within seconds and returns an annotated image with structured predictions, so you can immediately see how well it works. Adjust the detection classes and re-run until the results fit your scenario.
3. Like what you see?: With a free Roboflow account, you can deploy this workflow with the Serverless Cloud API, customize it in your workspace, or train RF-DETR (instance segmentation) on your own images for production accuracy.
4. Get started for free: No code or training data required to begin: upload, review the results, and make it part of your stack in a single click.