Typical Image retrieval systems evaluate the performance
using the average precision (AP) measure computed as the
area under the precision-recall curve for a query image. For
a set of test images, the mean Average Precision (mAP)
evaluates the overall performance.
We are interested in Precision at Rank-1, since the top retrieved
image is considered as the best match in the database
and acts as our annotation source. Hence, we choose to
evaluate the performance of our mobile app with a dierent
measure. We use the Precision at Rank-1 to say whether a
query image has been successfully annotated or not. We call
this as the Annotation Accuracy(AA).
In order to compute Annotation Accuracy for our Heritage
App, we collect test images of N buildings and structures using
mobile phone cameras at a particular site. This is our
test dataset. Annotation Accuracy is computed for each of
the N monuments and averaged to give the mean Annotation
Accuracy(mAA), which evaluates the performance of
our Heritage App for a specic site. Our methods of optimization
may not be applicable for a generic search.