waldofinder

Upload an image to start.

Benchmark metrics

Per-model metrics for reviewed presence decisions
Model	TP / FP / TN / FN	Recall Sensitivity · TPR	Specificity TNR	Precision	Accuracy	F1	Scored

Definitions & more metrics

Additional per-model classification metrics and excluded decisions
Model	NPV	False positive rate	False negative rate	Balanced accuracy	Errors	Abstentions	Unreviewed	Unsure reviews

Ground truth is your Present/Absent review of the input image, saved per model and region. The location marker alone does not label presence. Present/Absent predictions below the run’s confidence threshold and uncertain answers abstain. Errors, abstentions and missing/unsure reviews are excluded from rates. A dash means no eligible denominator. These measure presence in each input: a crop for grid/BSP or the full image for direct bounding boxes. BSP parent and overlapping crops count separately. They do not measure exact localization, verified identity or scene-level search success.

Sensitivity = Recall = TPR: TP / (TP + FN)
Specificity = TNR: TN / (TN + FP)
Precision: TP / (TP + FP)
Accuracy: (TP + TN) / scored decisions
F1: 2TP / (2TP + FP + FN)
Negative predictive value (NPV): TN / (TN + FN)

Benchmark metrics

Search configuration

Review region

Run history