Target Detection models
Target Detection scores each pixel for one selected target class. It is intended for finding a particular material against examples of the expected background.
Requirements
- A finalized dataset version with at least two classes.
- Exactly one Target class.
- One or more Non-target Classes.
When you choose the target, the other selected dataset classes provide the non-target examples. Include representative backgrounds and confusers; a narrow non-target set can produce convincing scores on unfamiliar material.
Evaluation
Review precision, recall or true-positive rate, false-positive rate, F1, AUROC, and AUPRC. Choose a threshold using the operational cost of missed targets and false alarms, not a universal confidence value.