Evaluation Metrics
Accuracy
- Accuracy : is the percentage of correctly predicted examples out of all predictions, formally known as
\(Accuracy = \frac{TP + TN}{TP + FP + TN + FN}\)
Precision
- Precision is the the probability of the predicted bounding boxes matching actual ground truth boxes, also referred to as the positive predictive value.
\(Precision = \frac{TP}{TP + FP} = \frac{true \ object \ detection}{all \ detected \ boxes}\)
Recall
- Recall is the true positive rate, also referred to as sensitivity, measures the probability of ground truth objects being correctly detected.
\(Recall = \frac{TP}{TP + FN} = \frac{true \ object \ detection}{all \ ground \ truth \ boxes}\)
AP - Average Precision
- it is a single number metric that encapsulates both precision and recall and summarizes the Precision-Recall curve by averaging precision across recall values from 0 to 1
\(AP = \dfrac{1}{11} \sum_{r \in \{0,0.1,0.2,...,1\}} p_{interp}(r)\)
where
\(p_{interp}(r) = \max p(\hat{r})\)
mAP (Mean Average Precision)
- count the accumulated TP and the accumulated FP and compute the precision/recall at each line. Average Precision is computed as the average precision at 11 equally spaced recall levels.