Lab notes · Emergency response
Detecting fire, smoke and vehicles in one pass
No off-the-shelf model detects cars, trucks, fire and smoke. Standard detectors know vehicles but not fire. Fire models know flame but not vehicles.
So we built the other half: several models run on every frame, and their results are merged into one set of labels. A 19-second 4K aerial clip of a trailer fire gave us 2,886 detections and a frame-by-frame data file you can search.
The biggest improvement did not come from the model. It came from the resolution we ran it at. At 1536 pixels the flame was found in 23% of frames. At 1920 pixels, in 65%.
This was a capability demo, not a client project. The same approach works for incident review, site safety, infrastructure inspection and security video that nobody has time to watch.
Before you train a new model, check your inference resolution. It can be the cheapest big win.
Capability demo built by IPD Scientific. More lab notes →
Have footage like this?
Book a free 15-minute call, or email us a sample image or short clip. We'll tell you honestly whether computer vision can do the job.