Closed MateoLostanlen closed 9 months ago
We have a bandit error because of try catch in loop for update_live_alerts_data call back I propose to manage it out of this PR
hello @Akilditu ,
Pyro-engine combines 4 frames to send an alert in order to reduce FP rate. The box we display is the prediction of the model on each frame, if the model do not detect anything there is no box
@Akilditu I understand your concern about the consistency of the detection flow. However, it's important to consider that every frame, whether it has a detected box or not, contains valuable information about the presence or absence of fire at a given moment. Removing frames without detections could potentially discard subtle cues or indicators that are crucial for a comprehensive understanding of the fire's evolution.
Firefighters rely on the full context of an evolving situation. Even frames where the AI does not detect fire can serve as negative data points that contribute to the overall pattern of the event. These 'empty' frames can act as confirmation that the fire has not spread to certain areas at specific times, which is just as important as knowing where the fire is active.
Therefore, keeping all frames ensures that firefighters get the complete picture, which is essential for their safety and the effectiveness of their response. It is better to err on the side of providing too much information rather than too little in such critical applications.
The PR to display the prediction the firefighters wanted!
I've added a button to toggle the prediction on or off for a better view in the case of a small fire.
I've removed the question "Is the alert relevant?", which is redundant with the question below, to free up space.
I'm going to update the devices with the latest version of engine so that they all send the prediction.