Temporal Smoke Verification
-
Pick a sequence
Choose one of the real camera sequences — two early wildfires and two false-positive look-alikes (clouds, haze).
-
Watch it play
The sequence loops automatically. Each box is a tracked candidate ("tube"), colored by the verdict — orange means judged smoke, gray means rejected.
-
See what the classifier sees
Below the animation, each tube's stabilized crops are laid out over time — the background holds still, so the only thing moving is the candidate.
-
Run the model live (optional)
Re-run the full pipeline from scratch on the Space's hardware to verify the precomputed results.
Overview
Temporal Smoke Verification tackles the hardest part of automated wildfire detection: telling real smoke from things that merely look like it. A single frame of drifting cloud, morning haze, or fog can fool a detector. This model watches a candidate over a sequence of frames instead — because genuine wildfire smoke behaves in ways a cloud doesn’t.
Built on Pyronear’s temporal model, it complements early detection by adding a second, time-aware opinion that filters out false alarms before they reach a responder.
How It Works
From a flagged candidate to a verdict you can see:
Tracked candidates
Each potential smoke is tracked across the sequence as a "tube" — a candidate followed frame by frame rather than judged in isolation.
Verdict over time
The classifier weighs how the candidate evolves and labels each tube — orange for judged smoke, gray for rejected look-alikes.
See what it sees
Stabilized crops of each tube are laid out over time, holding the background still so the only thing moving is the candidate itself.
Verify it live
Re-run the full pipeline from scratch on the Space's hardware to confirm the precomputed results for yourself.
Why Temporal Verification
Fewer false alarms
Clouds, haze, and fog are the classic triggers of false positives — watching over time filters them out.
Trust in alerts
Every false alarm erodes confidence and wastes responder time. Cleaner alerts keep people acting on the ones that matter.
A second opinion
It layers on top of frame-by-frame detection, adding temporal context without slowing the first warning.
Learn more about the project
See the wider early forest fire detection project and Pyronear's work — the detection system this verification step builds on.
View the project