Plainsight Detection Explorer

1. PROJECT SUMMARY
Detections explorer

2. PROBLEMS
We need to provide better ways for ML engineers to diagnose their models. Loss, confusion matrix, and P/R/F1 tables are reliable in most scenarios, however, viewing the actual detections and modifying them to improve the model are very desirable.

3. SOLUTIONS
Through competitive research and internal interviews, we developed an advanced filtering grid of detections for both streaming and bucket assets. Users can review all, narrow down to a specific time frame, depending on use-case, and send those images to either another colleague to review, or directly to retraining.