Skip to main content
Version: Next

Feedback

EVA analyzes situations and provides alarms based on detection scenarios using VLM.

Providing feedback on the alarms generated during this process is a critical step in continuously optimizing the performance of the EVA system.



Feedback Mechanism​

When an alarm is triggered, users can provide the following feedback on the alarm.

Users provide feedback on the message, image, and situational description determined by EVA based on the detection scenario.

  • Correct/Incorrect Judgment: Indicate whether the alarm is correct (True Positive) or incorrect (False Alarm).
  • Detailed Feedback: If incorrect, provide a detailed explanation of why it was wrong to help the VLM understand the context.


Utilizing Multi-modal Vector Store​

The feedback provided by users is stored in a Multi-modal Vector Store along with related information such as image snapshots, detection scenarios, and messages.

When the VLM needs to evaluate a similar situation in the future, it references these stored feedback data, including past success and failure cases. This significantly improves the accuracy of situation judgment.


💡 In the early stages of camera registration, false alarms may occur due to a lack of environment-specific information. However, as users consistently provide feedback on false alarms, EVA quickly adapts to the specific environment. Feedback on accurately detected situations serves as valuable training data for updating and improving the EVA model in the future.


In conclusion, the more feedback you provide, the more optimized EVA becomes for your operational environment, reducing unnecessary alarms and delivering only the necessary ones with greater accuracy.