safety-critical-apps

This paper proposes a knowledge distillation framework to compress vision-language models for on-device fire detection, showing that compact models can retain most of their teacher's capability while being deployable on embedded hardware.