Post-processing for object detection models
Model version:
A postprocessing layer that allows you to track bounding boxes across inference runs,
turning raw bounding boxes into stable “tracked” detections with unique identifiers.
This can also significantly reduce jitter and provide continuity of object labels across frames.
Parameters
Keep grace should be a number
Max. observations should be a number
Threshold should be a number
Limit objects by class should not be empty
Learn block thresholds
Score threshold
0.5
