Self-supervised pretraining with DenseMAE produces sub-megabyte models that detect wildfires from single-band MWIR imagery at 0.699 AP and 0.744 Fire-F1 while running under 150 ms on Jetson Xavier NX.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
On-Orbit Real-Time Wildfire Detection Under On-Board Constraints
Self-supervised pretraining with DenseMAE produces sub-megabyte models that detect wildfires from single-band MWIR imagery at 0.699 AP and 0.744 Fire-F1 while running under 150 ms on Jetson Xavier NX.