Racing Models, Not Opinions: How We Ran Wildfire ML R&D for Pyronear
How a literature survey, self-contained experiments, and a shared leaderboard turned 28 research papers into a production smoke verifier with 4x fewer false alarms.
How a literature survey, self-contained experiments, and a shared leaderboard turned 28 research papers into a production smoke verifier with 4x fewer false alarms.
How the temporal smoke verifier works, step by step — detection boxes become tubes, stabilized crops make motion legible, and a vision transformer plus a tiny temporal head learn that smoke grows and drifts.