Table 1.

Definitions for various retrieval (multiclass) evaluation metrics. Each metric is computed for the top-k images returned by the retrieval systems.

	Definition
Accuracy@k	$f(k)=\frac{1}{N_{classes}}\sum _{i=1}^{N_{classes}}\frac{tp+tn}{tp+fp+fn+tn}$
Precision@k	$f(k)=\frac{1}{N_{classes}}\sum _{i=1}^{N_{classes}}\frac{tp}{tp+fp}$
Recall@k	$f(k)=\frac{1}{N_{classes}}\sum _{i=1}^{N_{classes}}\frac{tp}{tp+fn}$
F1 – score@k	$f(k)=\frac{2\cdot Precision\cdot Recall}{Precision+Recall}$
Retrieval Score	$s=\sum _{k=1}^{n}f(k)\cdot {\left(\frac{1}{2}\right)}^k$

tp := true positives, tn := true negatives

fp := false positives, fn := false negatives