ACCURACY,PRECISION, SENSITIVITY/RECALL

• ACCURACY [Õigsus] = (TN+TP) / (TP+TN+FN+FP)
  • Correct predictions out of all
• PRECISION [Täpsus] = TP / (TP + FP) - How many of my positive classifications were actually positive
• SENSITIVIT/RECALL [Saagis] = TP/(TP+FN) - How many of the positives was I able to accurately classify as positive.

TP,FP,FN,TN - CONFUSION MATRIX

PRECISION-RECALL CURVE / AUC

Instance segmentation evaluation criteria

In example we have 5 images